From aab385d01b4311726127397552d791f4d71b7147 Mon Sep 17 00:00:00 2001 From: w-e-w <40751091+w-e-w@users.noreply.github.com> Date: Sun, 3 Sep 2023 11:56:02 +0900 Subject: [PATCH 001/139] thread safe extra network list_items --- extensions-builtin/Lora/ui_extra_networks_lora.py | 10 +++++----- modules/ui_extra_networks.py | 2 ++ modules/ui_extra_networks_checkpoints.py | 6 +++--- modules/ui_extra_networks_hypernets.py | 5 +++-- modules/ui_extra_networks_textual_inversion.py | 5 +++-- 5 files changed, 16 insertions(+), 12 deletions(-) diff --git a/extensions-builtin/Lora/ui_extra_networks_lora.py b/extensions-builtin/Lora/ui_extra_networks_lora.py index 55409a782..e9f300621 100644 --- a/extensions-builtin/Lora/ui_extra_networks_lora.py +++ b/extensions-builtin/Lora/ui_extra_networks_lora.py @@ -66,11 +66,11 @@ class ExtraNetworksPageLora(ui_extra_networks.ExtraNetworksPage): return item def list_items(self): - for index, name in enumerate(networks.available_networks): - item = self.create_item(name, index) - - if item is not None: - yield item + with self.thread_lock: + for index, name in enumerate(networks.available_networks): + item = self.create_item(name, index) + if item is not None: + yield item def allowed_directories_for_previews(self): return [shared.cmd_opts.lora_dir, shared.cmd_opts.lyco_dir_backcompat] diff --git a/modules/ui_extra_networks.py b/modules/ui_extra_networks.py index 063bd7b80..564bab7fe 100644 --- a/modules/ui_extra_networks.py +++ b/modules/ui_extra_networks.py @@ -1,6 +1,7 @@ import os.path import urllib.parse from pathlib import Path +from threading import Lock from modules import shared, ui_extra_networks_user_metadata, errors, extra_networks from modules.images import read_info_from_image, save_image_with_geninfo @@ -94,6 +95,7 @@ class ExtraNetworksPage: self.allow_negative_prompt = False self.metadata = {} self.items = {} + self.thread_lock = Lock() def refresh(self): pass diff --git a/modules/ui_extra_networks_checkpoints.py b/modules/ui_extra_networks_checkpoints.py index ca6c26076..2753214fa 100644 --- a/modules/ui_extra_networks_checkpoints.py +++ b/modules/ui_extra_networks_checkpoints.py @@ -30,9 +30,9 @@ class ExtraNetworksPageCheckpoints(ui_extra_networks.ExtraNetworksPage): } def list_items(self): - names = list(sd_models.checkpoints_list) - for index, name in enumerate(names): - yield self.create_item(name, index) + with self.thread_lock: + for index, name in enumerate(sd_models.checkpoints_list): + yield self.create_item(name, index) def allowed_directories_for_previews(self): return [v for v in [shared.cmd_opts.ckpt_dir, sd_models.model_path] if v is not None] diff --git a/modules/ui_extra_networks_hypernets.py b/modules/ui_extra_networks_hypernets.py index 4cedf0851..411b4f111 100644 --- a/modules/ui_extra_networks_hypernets.py +++ b/modules/ui_extra_networks_hypernets.py @@ -31,8 +31,9 @@ class ExtraNetworksPageHypernetworks(ui_extra_networks.ExtraNetworksPage): } def list_items(self): - for index, name in enumerate(shared.hypernetworks): - yield self.create_item(name, index) + with self.thread_lock: + for index, name in enumerate(shared.hypernetworks): + yield self.create_item(name, index) def allowed_directories_for_previews(self): return [shared.cmd_opts.hypernetwork_dir] diff --git a/modules/ui_extra_networks_textual_inversion.py b/modules/ui_extra_networks_textual_inversion.py index 55ef0ea7b..d25b45d61 100644 --- a/modules/ui_extra_networks_textual_inversion.py +++ b/modules/ui_extra_networks_textual_inversion.py @@ -29,8 +29,9 @@ class ExtraNetworksPageTextualInversion(ui_extra_networks.ExtraNetworksPage): } def list_items(self): - for index, name in enumerate(sd_hijack.model_hijack.embedding_db.word_embeddings): - yield self.create_item(name, index) + with self.thread_lock: + for index, name in enumerate(sd_hijack.model_hijack.embedding_db.word_embeddings): + yield self.create_item(name, index) def allowed_directories_for_previews(self): return list(sd_hijack.model_hijack.embedding_db.embedding_dirs) From 25de9a785cc9e93c16626db6ab5b16824443de53 Mon Sep 17 00:00:00 2001 From: w-e-w <40751091+w-e-w@users.noreply.github.com> Date: Sat, 9 Sep 2023 16:56:19 +0900 Subject: [PATCH 002/139] Revert "thread safe extra network list_items" This reverts commit aab385d01b4311726127397552d791f4d71b7147. --- extensions-builtin/Lora/ui_extra_networks_lora.py | 10 +++++----- modules/ui_extra_networks.py | 2 -- modules/ui_extra_networks_checkpoints.py | 6 +++--- modules/ui_extra_networks_hypernets.py | 5 ++--- modules/ui_extra_networks_textual_inversion.py | 5 ++--- 5 files changed, 12 insertions(+), 16 deletions(-) diff --git a/extensions-builtin/Lora/ui_extra_networks_lora.py b/extensions-builtin/Lora/ui_extra_networks_lora.py index e9f300621..55409a782 100644 --- a/extensions-builtin/Lora/ui_extra_networks_lora.py +++ b/extensions-builtin/Lora/ui_extra_networks_lora.py @@ -66,11 +66,11 @@ class ExtraNetworksPageLora(ui_extra_networks.ExtraNetworksPage): return item def list_items(self): - with self.thread_lock: - for index, name in enumerate(networks.available_networks): - item = self.create_item(name, index) - if item is not None: - yield item + for index, name in enumerate(networks.available_networks): + item = self.create_item(name, index) + + if item is not None: + yield item def allowed_directories_for_previews(self): return [shared.cmd_opts.lora_dir, shared.cmd_opts.lyco_dir_backcompat] diff --git a/modules/ui_extra_networks.py b/modules/ui_extra_networks.py index 564bab7fe..063bd7b80 100644 --- a/modules/ui_extra_networks.py +++ b/modules/ui_extra_networks.py @@ -1,7 +1,6 @@ import os.path import urllib.parse from pathlib import Path -from threading import Lock from modules import shared, ui_extra_networks_user_metadata, errors, extra_networks from modules.images import read_info_from_image, save_image_with_geninfo @@ -95,7 +94,6 @@ class ExtraNetworksPage: self.allow_negative_prompt = False self.metadata = {} self.items = {} - self.thread_lock = Lock() def refresh(self): pass diff --git a/modules/ui_extra_networks_checkpoints.py b/modules/ui_extra_networks_checkpoints.py index 2753214fa..ca6c26076 100644 --- a/modules/ui_extra_networks_checkpoints.py +++ b/modules/ui_extra_networks_checkpoints.py @@ -30,9 +30,9 @@ class ExtraNetworksPageCheckpoints(ui_extra_networks.ExtraNetworksPage): } def list_items(self): - with self.thread_lock: - for index, name in enumerate(sd_models.checkpoints_list): - yield self.create_item(name, index) + names = list(sd_models.checkpoints_list) + for index, name in enumerate(names): + yield self.create_item(name, index) def allowed_directories_for_previews(self): return [v for v in [shared.cmd_opts.ckpt_dir, sd_models.model_path] if v is not None] diff --git a/modules/ui_extra_networks_hypernets.py b/modules/ui_extra_networks_hypernets.py index 411b4f111..4cedf0851 100644 --- a/modules/ui_extra_networks_hypernets.py +++ b/modules/ui_extra_networks_hypernets.py @@ -31,9 +31,8 @@ class ExtraNetworksPageHypernetworks(ui_extra_networks.ExtraNetworksPage): } def list_items(self): - with self.thread_lock: - for index, name in enumerate(shared.hypernetworks): - yield self.create_item(name, index) + for index, name in enumerate(shared.hypernetworks): + yield self.create_item(name, index) def allowed_directories_for_previews(self): return [shared.cmd_opts.hypernetwork_dir] diff --git a/modules/ui_extra_networks_textual_inversion.py b/modules/ui_extra_networks_textual_inversion.py index d25b45d61..55ef0ea7b 100644 --- a/modules/ui_extra_networks_textual_inversion.py +++ b/modules/ui_extra_networks_textual_inversion.py @@ -29,9 +29,8 @@ class ExtraNetworksPageTextualInversion(ui_extra_networks.ExtraNetworksPage): } def list_items(self): - with self.thread_lock: - for index, name in enumerate(sd_hijack.model_hijack.embedding_db.word_embeddings): - yield self.create_item(name, index) + for index, name in enumerate(sd_hijack.model_hijack.embedding_db.word_embeddings): + yield self.create_item(name, index) def allowed_directories_for_previews(self): return list(sd_hijack.model_hijack.embedding_db.embedding_dirs) From f5959c1c3022c454de22fab749d0f06ab3219868 Mon Sep 17 00:00:00 2001 From: w-e-w <40751091+w-e-w@users.noreply.github.com> Date: Sat, 9 Sep 2023 17:05:50 +0900 Subject: [PATCH 003/139] thread safe extra network using list --- extensions-builtin/Lora/ui_extra_networks_lora.py | 3 ++- modules/ui_extra_networks_hypernets.py | 3 ++- modules/ui_extra_networks_textual_inversion.py | 3 ++- 3 files changed, 6 insertions(+), 3 deletions(-) diff --git a/extensions-builtin/Lora/ui_extra_networks_lora.py b/extensions-builtin/Lora/ui_extra_networks_lora.py index 55409a782..e74daa770 100644 --- a/extensions-builtin/Lora/ui_extra_networks_lora.py +++ b/extensions-builtin/Lora/ui_extra_networks_lora.py @@ -66,7 +66,8 @@ class ExtraNetworksPageLora(ui_extra_networks.ExtraNetworksPage): return item def list_items(self): - for index, name in enumerate(networks.available_networks): + names = list(networks.available_networks) + for index, name in enumerate(names): item = self.create_item(name, index) if item is not None: diff --git a/modules/ui_extra_networks_hypernets.py b/modules/ui_extra_networks_hypernets.py index 4cedf0851..5f5904915 100644 --- a/modules/ui_extra_networks_hypernets.py +++ b/modules/ui_extra_networks_hypernets.py @@ -31,7 +31,8 @@ class ExtraNetworksPageHypernetworks(ui_extra_networks.ExtraNetworksPage): } def list_items(self): - for index, name in enumerate(shared.hypernetworks): + names = list(shared.hypernetworks) + for index, name in enumerate(names): yield self.create_item(name, index) def allowed_directories_for_previews(self): diff --git a/modules/ui_extra_networks_textual_inversion.py b/modules/ui_extra_networks_textual_inversion.py index 55ef0ea7b..40ab0aca3 100644 --- a/modules/ui_extra_networks_textual_inversion.py +++ b/modules/ui_extra_networks_textual_inversion.py @@ -29,7 +29,8 @@ class ExtraNetworksPageTextualInversion(ui_extra_networks.ExtraNetworksPage): } def list_items(self): - for index, name in enumerate(sd_hijack.model_hijack.embedding_db.word_embeddings): + names = list(sd_hijack.model_hijack.embedding_db.word_embeddings) + for index, name in enumerate(names): yield self.create_item(name, index) def allowed_directories_for_previews(self): From e785402b6acca12108e15224ff80d58817ab3c27 Mon Sep 17 00:00:00 2001 From: w-e-w <40751091+w-e-w@users.noreply.github.com> Date: Sat, 9 Sep 2023 17:28:06 +0900 Subject: [PATCH 004/139] return nothing if not found --- extensions-builtin/Lora/ui_extra_networks_lora.py | 3 ++- modules/ui_extra_networks_checkpoints.py | 7 ++++++- modules/ui_extra_networks_hypernets.py | 9 +++++++-- modules/ui_extra_networks_textual_inversion.py | 6 +++++- 4 files changed, 20 insertions(+), 5 deletions(-) diff --git a/extensions-builtin/Lora/ui_extra_networks_lora.py b/extensions-builtin/Lora/ui_extra_networks_lora.py index e74daa770..dac90a86c 100644 --- a/extensions-builtin/Lora/ui_extra_networks_lora.py +++ b/extensions-builtin/Lora/ui_extra_networks_lora.py @@ -17,6 +17,8 @@ class ExtraNetworksPageLora(ui_extra_networks.ExtraNetworksPage): def create_item(self, name, index=None, enable_filter=True): lora_on_disk = networks.available_networks.get(name) + if lora_on_disk is None: + return path, ext = os.path.splitext(lora_on_disk.filename) @@ -69,7 +71,6 @@ class ExtraNetworksPageLora(ui_extra_networks.ExtraNetworksPage): names = list(networks.available_networks) for index, name in enumerate(names): item = self.create_item(name, index) - if item is not None: yield item diff --git a/modules/ui_extra_networks_checkpoints.py b/modules/ui_extra_networks_checkpoints.py index ca6c26076..35e958a00 100644 --- a/modules/ui_extra_networks_checkpoints.py +++ b/modules/ui_extra_networks_checkpoints.py @@ -15,6 +15,9 @@ class ExtraNetworksPageCheckpoints(ui_extra_networks.ExtraNetworksPage): def create_item(self, name, index=None, enable_filter=True): checkpoint: sd_models.CheckpointInfo = sd_models.checkpoint_aliases.get(name) + if checkpoint is None: + return + path, ext = os.path.splitext(checkpoint.filename) return { "name": checkpoint.name_for_extra, @@ -32,7 +35,9 @@ class ExtraNetworksPageCheckpoints(ui_extra_networks.ExtraNetworksPage): def list_items(self): names = list(sd_models.checkpoints_list) for index, name in enumerate(names): - yield self.create_item(name, index) + item = self.create_item(name, index) + if item is not None: + yield item def allowed_directories_for_previews(self): return [v for v in [shared.cmd_opts.ckpt_dir, sd_models.model_path] if v is not None] diff --git a/modules/ui_extra_networks_hypernets.py b/modules/ui_extra_networks_hypernets.py index 5f5904915..74f7d8472 100644 --- a/modules/ui_extra_networks_hypernets.py +++ b/modules/ui_extra_networks_hypernets.py @@ -13,7 +13,10 @@ class ExtraNetworksPageHypernetworks(ui_extra_networks.ExtraNetworksPage): shared.reload_hypernetworks() def create_item(self, name, index=None, enable_filter=True): - full_path = shared.hypernetworks[name] + full_path = shared.hypernetworks.get(name) + if full_path is None: + return + path, ext = os.path.splitext(full_path) sha256 = sha256_from_cache(full_path, f'hypernet/{name}') shorthash = sha256[0:10] if sha256 else None @@ -33,7 +36,9 @@ class ExtraNetworksPageHypernetworks(ui_extra_networks.ExtraNetworksPage): def list_items(self): names = list(shared.hypernetworks) for index, name in enumerate(names): - yield self.create_item(name, index) + item = self.create_item(name, index) + if item is not None: + yield item def allowed_directories_for_previews(self): return [shared.cmd_opts.hypernetwork_dir] diff --git a/modules/ui_extra_networks_textual_inversion.py b/modules/ui_extra_networks_textual_inversion.py index 40ab0aca3..71c38fabc 100644 --- a/modules/ui_extra_networks_textual_inversion.py +++ b/modules/ui_extra_networks_textual_inversion.py @@ -14,6 +14,8 @@ class ExtraNetworksPageTextualInversion(ui_extra_networks.ExtraNetworksPage): def create_item(self, name, index=None, enable_filter=True): embedding = sd_hijack.model_hijack.embedding_db.word_embeddings.get(name) + if embedding is None: + return path, ext = os.path.splitext(embedding.filename) return { @@ -31,7 +33,9 @@ class ExtraNetworksPageTextualInversion(ui_extra_networks.ExtraNetworksPage): def list_items(self): names = list(sd_hijack.model_hijack.embedding_db.word_embeddings) for index, name in enumerate(names): - yield self.create_item(name, index) + item = self.create_item(name, index) + if item is not None: + yield item def allowed_directories_for_previews(self): return list(sd_hijack.model_hijack.embedding_db.embedding_dirs) From 74b80e72115af46bf1c04167a30f9ec5025cb464 Mon Sep 17 00:00:00 2001 From: w-e-w <40751091+w-e-w@users.noreply.github.com> Date: Tue, 12 Sep 2023 09:29:07 +0900 Subject: [PATCH 005/139] add comment --- extensions-builtin/Lora/ui_extra_networks_lora.py | 1 + modules/ui_extra_networks_checkpoints.py | 1 + modules/ui_extra_networks_hypernets.py | 1 + modules/ui_extra_networks_textual_inversion.py | 1 + 4 files changed, 4 insertions(+) diff --git a/extensions-builtin/Lora/ui_extra_networks_lora.py b/extensions-builtin/Lora/ui_extra_networks_lora.py index dac90a86c..df02c663b 100644 --- a/extensions-builtin/Lora/ui_extra_networks_lora.py +++ b/extensions-builtin/Lora/ui_extra_networks_lora.py @@ -68,6 +68,7 @@ class ExtraNetworksPageLora(ui_extra_networks.ExtraNetworksPage): return item def list_items(self): + # instantiate a list to protect against concurrent modification names = list(networks.available_networks) for index, name in enumerate(names): item = self.create_item(name, index) diff --git a/modules/ui_extra_networks_checkpoints.py b/modules/ui_extra_networks_checkpoints.py index 35e958a00..df7efb2e1 100644 --- a/modules/ui_extra_networks_checkpoints.py +++ b/modules/ui_extra_networks_checkpoints.py @@ -33,6 +33,7 @@ class ExtraNetworksPageCheckpoints(ui_extra_networks.ExtraNetworksPage): } def list_items(self): + # instantiate a list to protect against concurrent modification names = list(sd_models.checkpoints_list) for index, name in enumerate(names): item = self.create_item(name, index) diff --git a/modules/ui_extra_networks_hypernets.py b/modules/ui_extra_networks_hypernets.py index 74f7d8472..c96c4fa3b 100644 --- a/modules/ui_extra_networks_hypernets.py +++ b/modules/ui_extra_networks_hypernets.py @@ -34,6 +34,7 @@ class ExtraNetworksPageHypernetworks(ui_extra_networks.ExtraNetworksPage): } def list_items(self): + # instantiate a list to protect against concurrent modification names = list(shared.hypernetworks) for index, name in enumerate(names): item = self.create_item(name, index) diff --git a/modules/ui_extra_networks_textual_inversion.py b/modules/ui_extra_networks_textual_inversion.py index 71c38fabc..1b334fda1 100644 --- a/modules/ui_extra_networks_textual_inversion.py +++ b/modules/ui_extra_networks_textual_inversion.py @@ -31,6 +31,7 @@ class ExtraNetworksPageTextualInversion(ui_extra_networks.ExtraNetworksPage): } def list_items(self): + # instantiate a list to protect against concurrent modification names = list(sd_hijack.model_hijack.embedding_db.word_embeddings) for index, name in enumerate(names): item = self.create_item(name, index) From ec718f76b58b183859ed732e11ec748c41a13f76 Mon Sep 17 00:00:00 2001 From: v0xie <28695009+v0xie@users.noreply.github.com> Date: Tue, 17 Oct 2023 23:35:50 -0700 Subject: [PATCH 006/139] wip incorrect OFT implementation --- extensions-builtin/Lora/network_oft.py | 82 ++++++++++++++++++++++++++ extensions-builtin/Lora/networks.py | 5 ++ 2 files changed, 87 insertions(+) create mode 100644 extensions-builtin/Lora/network_oft.py diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py new file mode 100644 index 000000000..9ddb175ce --- /dev/null +++ b/extensions-builtin/Lora/network_oft.py @@ -0,0 +1,82 @@ +import torch +import network + + +class ModuleTypeOFT(network.ModuleType): + def create_module(self, net: network.Network, weights: network.NetworkWeights): + if all(x in weights.w for x in ["oft_blocks"]): + return NetworkModuleOFT(net, weights) + + return None + +# adapted from https://github.com/kohya-ss/sd-scripts/blob/main/networks/oft.py +class NetworkModuleOFT(network.NetworkModule): + def __init__(self, net: network.Network, weights: network.NetworkWeights): + super().__init__(net, weights) + + self.oft_blocks = weights.w["oft_blocks"] + self.alpha = weights.w["alpha"] + + self.dim = self.oft_blocks.shape[0] + self.num_blocks = self.dim + + #if type(self.alpha) == torch.Tensor: + # self.alpha = self.alpha.detach().numpy() + + if "Linear" in self.sd_module.__class__.__name__: + self.out_dim = self.sd_module.out_features + elif "Conv" in self.sd_module.__class__.__name__: + self.out_dim = self.sd_module.out_channels + + self.constraint = self.alpha * self.out_dim + self.block_size = self.out_dim // self.num_blocks + + self.oft_multiplier = self.multiplier() + + # replace forward method of original linear rather than replacing the module + # self.org_forward = self.sd_module.forward + # self.sd_module.forward = self.forward + + def get_weight(self): + block_Q = self.oft_blocks - self.oft_blocks.transpose(1, 2) + norm_Q = torch.norm(block_Q.flatten()) + new_norm_Q = torch.clamp(norm_Q, max=self.constraint) + block_Q = block_Q * ((new_norm_Q + 1e-8) / (norm_Q + 1e-8)) + I = torch.eye(self.block_size, device=self.oft_blocks.device).unsqueeze(0).repeat(self.num_blocks, 1, 1) + block_R = torch.matmul(I + block_Q, (I - block_Q).inverse()) + + block_R_weighted = self.oft_multiplier * block_R + (1 - self.oft_multiplier) * I + R = torch.block_diag(*block_R_weighted) + + return R + + def calc_updown(self, orig_weight): + oft_blocks = self.oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) + block_Q = oft_blocks - oft_blocks.transpose(1, 2) + norm_Q = torch.norm(block_Q.flatten()) + new_norm_Q = torch.clamp(norm_Q, max=self.constraint) + block_Q = block_Q * ((new_norm_Q + 1e-8) / (norm_Q + 1e-8)) + I = torch.eye(self.block_size, device=oft_blocks.device).unsqueeze(0).repeat(self.num_blocks, 1, 1) + block_R = torch.matmul(I + block_Q, (I - block_Q).inverse()) + + block_R_weighted = self.oft_multiplier * block_R + (1 - self.oft_multiplier) * I + R = torch.block_diag(*block_R_weighted) + #R = self.get_weight().to(orig_weight.device, dtype=orig_weight.dtype) + # W = R*W_0 + updown = orig_weight + R + output_shape = [R.size(0), orig_weight.size(1)] + return self.finalize_updown(updown, orig_weight, output_shape) + + # def forward(self, x, y=None): + # x = self.org_forward(x) + # if self.oft_multiplier == 0.0: + # return x + + # R = self.get_weight().to(x.device, dtype=x.dtype) + # if x.dim() == 4: + # x = x.permute(0, 2, 3, 1) + # x = torch.matmul(x, R) + # x = x.permute(0, 3, 1, 2) + # else: + # x = torch.matmul(x, R) + # return x diff --git a/extensions-builtin/Lora/networks.py b/extensions-builtin/Lora/networks.py index 60d8dec4c..bd1f1b756 100644 --- a/extensions-builtin/Lora/networks.py +++ b/extensions-builtin/Lora/networks.py @@ -11,6 +11,7 @@ import network_ia3 import network_lokr import network_full import network_norm +import network_oft import torch from typing import Union @@ -28,6 +29,7 @@ module_types = [ network_full.ModuleTypeFull(), network_norm.ModuleTypeNorm(), network_glora.ModuleTypeGLora(), + network_oft.ModuleTypeOFT(), ] @@ -183,6 +185,9 @@ def load_network(name, network_on_disk): elif sd_module is None and "lora_te1_text_model" in key_network_without_network_parts: key = key_network_without_network_parts.replace("lora_te1_text_model", "0_transformer_text_model") sd_module = shared.sd_model.network_layer_mapping.get(key, None) + elif sd_module is None and "oft_unet" in key_network_without_network_parts: + key = key_network_without_network_parts.replace("oft_unet", "diffusion_model") + sd_module = shared.sd_model.network_layer_mapping.get(key, None) # some SD1 Loras also have correct compvis keys if sd_module is None: From 1c6efdbba774d603c592debaccd6f5ad827bd1b2 Mon Sep 17 00:00:00 2001 From: v0xie <28695009+v0xie@users.noreply.github.com> Date: Wed, 18 Oct 2023 04:16:01 -0700 Subject: [PATCH 007/139] inference working but SLOW --- extensions-builtin/Lora/network_oft.py | 73 +++++++++++++------------- extensions-builtin/Lora/networks.py | 42 +++++++++++++-- 2 files changed, 75 insertions(+), 40 deletions(-) diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py index 9ddb175ce..f085eca53 100644 --- a/extensions-builtin/Lora/network_oft.py +++ b/extensions-builtin/Lora/network_oft.py @@ -12,6 +12,7 @@ class ModuleTypeOFT(network.ModuleType): # adapted from https://github.com/kohya-ss/sd-scripts/blob/main/networks/oft.py class NetworkModuleOFT(network.NetworkModule): def __init__(self, net: network.Network, weights: network.NetworkWeights): + super().__init__(net, weights) self.oft_blocks = weights.w["oft_blocks"] @@ -20,24 +21,29 @@ class NetworkModuleOFT(network.NetworkModule): self.dim = self.oft_blocks.shape[0] self.num_blocks = self.dim - #if type(self.alpha) == torch.Tensor: - # self.alpha = self.alpha.detach().numpy() - if "Linear" in self.sd_module.__class__.__name__: self.out_dim = self.sd_module.out_features elif "Conv" in self.sd_module.__class__.__name__: self.out_dim = self.sd_module.out_channels - self.constraint = self.alpha * self.out_dim + self.constraint = self.alpha + #self.constraint = self.alpha * self.out_dim self.block_size = self.out_dim // self.num_blocks - self.oft_multiplier = self.multiplier() + self.org_module: list[torch.Module] = [self.sd_module] - # replace forward method of original linear rather than replacing the module - # self.org_forward = self.sd_module.forward - # self.sd_module.forward = self.forward + self.R = self.get_weight() + + self.apply_to() + + # replace forward method of original linear rather than replacing the module + def apply_to(self): + self.org_forward = self.org_module[0].forward + self.org_module[0].forward = self.forward - def get_weight(self): + def get_weight(self, multiplier=None): + if not multiplier: + multiplier = self.multiplier() block_Q = self.oft_blocks - self.oft_blocks.transpose(1, 2) norm_Q = torch.norm(block_Q.flatten()) new_norm_Q = torch.clamp(norm_Q, max=self.constraint) @@ -45,38 +51,31 @@ class NetworkModuleOFT(network.NetworkModule): I = torch.eye(self.block_size, device=self.oft_blocks.device).unsqueeze(0).repeat(self.num_blocks, 1, 1) block_R = torch.matmul(I + block_Q, (I - block_Q).inverse()) - block_R_weighted = self.oft_multiplier * block_R + (1 - self.oft_multiplier) * I + block_R_weighted = multiplier * block_R + (1 - multiplier) * I R = torch.block_diag(*block_R_weighted) return R def calc_updown(self, orig_weight): - oft_blocks = self.oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) - block_Q = oft_blocks - oft_blocks.transpose(1, 2) - norm_Q = torch.norm(block_Q.flatten()) - new_norm_Q = torch.clamp(norm_Q, max=self.constraint) - block_Q = block_Q * ((new_norm_Q + 1e-8) / (norm_Q + 1e-8)) - I = torch.eye(self.block_size, device=oft_blocks.device).unsqueeze(0).repeat(self.num_blocks, 1, 1) - block_R = torch.matmul(I + block_Q, (I - block_Q).inverse()) - - block_R_weighted = self.oft_multiplier * block_R + (1 - self.oft_multiplier) * I - R = torch.block_diag(*block_R_weighted) - #R = self.get_weight().to(orig_weight.device, dtype=orig_weight.dtype) - # W = R*W_0 - updown = orig_weight + R - output_shape = [R.size(0), orig_weight.size(1)] + R = self.R + if orig_weight.dim() == 4: + weight = torch.einsum("oihw, op -> pihw", orig_weight, R) + else: + weight = torch.einsum("oi, op -> pi", orig_weight, R) + updown = orig_weight @ R + output_shape = [orig_weight.size(0), R.size(1)] + #output_shape = [R.size(0), orig_weight.size(1)] return self.finalize_updown(updown, orig_weight, output_shape) - # def forward(self, x, y=None): - # x = self.org_forward(x) - # if self.oft_multiplier == 0.0: - # return x - - # R = self.get_weight().to(x.device, dtype=x.dtype) - # if x.dim() == 4: - # x = x.permute(0, 2, 3, 1) - # x = torch.matmul(x, R) - # x = x.permute(0, 3, 1, 2) - # else: - # x = torch.matmul(x, R) - # return x + def forward(self, x, y=None): + x = self.org_forward(x) + if self.multiplier() == 0.0: + return x + R = self.get_weight().to(x.device, dtype=x.dtype) + if x.dim() == 4: + x = x.permute(0, 2, 3, 1) + x = torch.matmul(x, R) + x = x.permute(0, 3, 1, 2) + else: + x = torch.matmul(x, R) + return x diff --git a/extensions-builtin/Lora/networks.py b/extensions-builtin/Lora/networks.py index bd1f1b756..e5e73450b 100644 --- a/extensions-builtin/Lora/networks.py +++ b/extensions-builtin/Lora/networks.py @@ -169,6 +169,10 @@ def load_network(name, network_on_disk): else: emb_dict[vec_name] = weight bundle_embeddings[emb_name] = emb_dict + + #if key_network_without_network_parts == "oft_unet": + # print(key_network_without_network_parts) + # pass key = convert_diffusers_name_to_compvis(key_network_without_network_parts, is_sd2) sd_module = shared.sd_model.network_layer_mapping.get(key, None) @@ -185,15 +189,39 @@ def load_network(name, network_on_disk): elif sd_module is None and "lora_te1_text_model" in key_network_without_network_parts: key = key_network_without_network_parts.replace("lora_te1_text_model", "0_transformer_text_model") sd_module = shared.sd_model.network_layer_mapping.get(key, None) - elif sd_module is None and "oft_unet" in key_network_without_network_parts: - key = key_network_without_network_parts.replace("oft_unet", "diffusion_model") - sd_module = shared.sd_model.network_layer_mapping.get(key, None) # some SD1 Loras also have correct compvis keys if sd_module is None: key = key_network_without_network_parts.replace("lora_te1_text_model", "transformer_text_model") sd_module = shared.sd_model.network_layer_mapping.get(key, None) + elif sd_module is None and "oft_unet" in key_network_without_network_parts: + # UNET_TARGET_REPLACE_MODULE_ALL_LINEAR = ["Transformer2DModel"] + # UNET_TARGET_REPLACE_MODULE_CONV2D_3X3 = ["ResnetBlock2D", "Downsample2D", "Upsample2D"] + UNET_TARGET_REPLACE_MODULE_ATTN_ONLY = ["CrossAttention"] + # TODO: Change matchedm odules based on whether all linear, conv, etc + + key = key_network_without_network_parts.replace("oft_unet", "diffusion_model") + sd_module = shared.sd_model.network_layer_mapping.get(key, None) + #key_no_suffix = key.rsplit("_to_", 1)[0] + ## Match all modules of class CrossAttention + #replace_module_list = [] + #for module_type in UNET_TARGET_REPLACE_MODULE_ATTN_ONLY: + # replace_module_list += [module for k, module in shared.sd_model.network_layer_mapping.items() if module_type in module.__class__.__name__] + + #matched_module = replace_module_list.get(key_no_suffix, None) + #if key.endswith('to_q'): + # sd_module = matched_module.to_q or None + #if key.endswith('to_k'): + # sd_module = matched_module.to_k or None + #if key.endswith('to_v'): + # sd_module = matched_module.to_v or None + #if key.endswith('to_out_0'): + # sd_module = matched_module.to_out[0] or None + #if key.endswith('to_out_1'): + # sd_module = matched_module.to_out[1] or None + + if sd_module is None: keys_failed_to_match[key_network] = key continue @@ -214,6 +242,14 @@ def load_network(name, network_on_disk): raise AssertionError(f"Could not find a module type (out of {', '.join([x.__class__.__name__ for x in module_types])}) that would accept those keys: {', '.join(weights.w)}") net.modules[key] = net_module + + # replaces forward method of original Linear + # applied_to_count = 0 + #for key, created_module in net.modules.items(): + # if isinstance(created_module, network_oft.NetworkModuleOFT): + # net_module.apply_to() + #applied_to_count += 1 + # print(f'Applied OFT modules: {applied_to_count}') embeddings = {} for emb_name, data in bundle_embeddings.items(): From 853e21d98eada4db9a9fd1ae8eda90cf763e2818 Mon Sep 17 00:00:00 2001 From: v0xie <28695009+v0xie@users.noreply.github.com> Date: Wed, 18 Oct 2023 04:27:44 -0700 Subject: [PATCH 008/139] faster by using cached R in forward --- extensions-builtin/Lora/network_oft.py | 17 ++++++++++++++--- 1 file changed, 14 insertions(+), 3 deletions(-) diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py index f085eca53..68efb1db9 100644 --- a/extensions-builtin/Lora/network_oft.py +++ b/extensions-builtin/Lora/network_oft.py @@ -57,21 +57,32 @@ class NetworkModuleOFT(network.NetworkModule): return R def calc_updown(self, orig_weight): + # this works R = self.R + + # this causes major deepfrying i.e. just doesn't work + # R = self.R.to(orig_weight.device, dtype=orig_weight.dtype) + if orig_weight.dim() == 4: weight = torch.einsum("oihw, op -> pihw", orig_weight, R) else: weight = torch.einsum("oi, op -> pi", orig_weight, R) + updown = orig_weight @ R - output_shape = [orig_weight.size(0), R.size(1)] - #output_shape = [R.size(0), orig_weight.size(1)] + output_shape = self.oft_blocks.shape + + ## this works + # updown = orig_weight @ R + # output_shape = [orig_weight.size(0), R.size(1)] + return self.finalize_updown(updown, orig_weight, output_shape) def forward(self, x, y=None): x = self.org_forward(x) if self.multiplier() == 0.0: return x - R = self.get_weight().to(x.device, dtype=x.dtype) + #R = self.get_weight().to(x.device, dtype=x.dtype) + R = self.R.to(x.device, dtype=x.dtype) if x.dim() == 4: x = x.permute(0, 2, 3, 1) x = torch.matmul(x, R) From eb01d7f0e0fb46285985803296a25715165fb3f9 Mon Sep 17 00:00:00 2001 From: v0xie <28695009+v0xie@users.noreply.github.com> Date: Wed, 18 Oct 2023 04:56:53 -0700 Subject: [PATCH 009/139] faster by calculating R in updown and using cached R in forward --- extensions-builtin/Lora/network_oft.py | 15 ++++++++------- 1 file changed, 8 insertions(+), 7 deletions(-) diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py index 68efb1db9..fd5b0c0fd 100644 --- a/extensions-builtin/Lora/network_oft.py +++ b/extensions-builtin/Lora/network_oft.py @@ -58,17 +58,18 @@ class NetworkModuleOFT(network.NetworkModule): def calc_updown(self, orig_weight): # this works - R = self.R + # R = self.R + self.R = self.get_weight(self.multiplier()) - # this causes major deepfrying i.e. just doesn't work + # sending R to device causes major deepfrying i.e. just doesn't work # R = self.R.to(orig_weight.device, dtype=orig_weight.dtype) - if orig_weight.dim() == 4: - weight = torch.einsum("oihw, op -> pihw", orig_weight, R) - else: - weight = torch.einsum("oi, op -> pi", orig_weight, R) + # if orig_weight.dim() == 4: + # weight = torch.einsum("oihw, op -> pihw", orig_weight, R) + # else: + # weight = torch.einsum("oi, op -> pi", orig_weight, R) - updown = orig_weight @ R + updown = orig_weight @ self.R output_shape = self.oft_blocks.shape ## this works From 321680ccd0e0404223fbdf4f26498f7d0317fb75 Mon Sep 17 00:00:00 2001 From: v0xie <28695009+v0xie@users.noreply.github.com> Date: Thu, 19 Oct 2023 12:41:17 -0700 Subject: [PATCH 010/139] refactor: fix constraint, re-use get_weight --- extensions-builtin/Lora/network_oft.py | 40 +++++++++++--------------- 1 file changed, 16 insertions(+), 24 deletions(-) diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py index fd5b0c0fd..2af1bc4cf 100644 --- a/extensions-builtin/Lora/network_oft.py +++ b/extensions-builtin/Lora/network_oft.py @@ -9,7 +9,7 @@ class ModuleTypeOFT(network.ModuleType): return None -# adapted from https://github.com/kohya-ss/sd-scripts/blob/main/networks/oft.py +# adapted from kohya's implementation https://github.com/kohya-ss/sd-scripts/blob/main/networks/oft.py class NetworkModuleOFT(network.NetworkModule): def __init__(self, net: network.Network, weights: network.NetworkWeights): @@ -17,7 +17,6 @@ class NetworkModuleOFT(network.NetworkModule): self.oft_blocks = weights.w["oft_blocks"] self.alpha = weights.w["alpha"] - self.dim = self.oft_blocks.shape[0] self.num_blocks = self.dim @@ -26,64 +25,57 @@ class NetworkModuleOFT(network.NetworkModule): elif "Conv" in self.sd_module.__class__.__name__: self.out_dim = self.sd_module.out_channels - self.constraint = self.alpha - #self.constraint = self.alpha * self.out_dim + self.constraint = self.alpha * self.out_dim self.block_size = self.out_dim // self.num_blocks self.org_module: list[torch.Module] = [self.sd_module] - - self.R = self.get_weight() - + self.R = self.get_weight(self.oft_blocks) self.apply_to() # replace forward method of original linear rather than replacing the module + # how do we revert this to unload the weights? def apply_to(self): self.org_forward = self.org_module[0].forward self.org_module[0].forward = self.forward - def get_weight(self, multiplier=None): - if not multiplier: - multiplier = self.multiplier() - block_Q = self.oft_blocks - self.oft_blocks.transpose(1, 2) + def get_weight(self, oft_blocks, multiplier=None): + block_Q = oft_blocks - oft_blocks.transpose(1, 2) norm_Q = torch.norm(block_Q.flatten()) new_norm_Q = torch.clamp(norm_Q, max=self.constraint) block_Q = block_Q * ((new_norm_Q + 1e-8) / (norm_Q + 1e-8)) I = torch.eye(self.block_size, device=self.oft_blocks.device).unsqueeze(0).repeat(self.num_blocks, 1, 1) block_R = torch.matmul(I + block_Q, (I - block_Q).inverse()) - - block_R_weighted = multiplier * block_R + (1 - multiplier) * I - R = torch.block_diag(*block_R_weighted) + #block_R_weighted = multiplier * block_R + (1 - multiplier) * I + #R = torch.block_diag(*block_R_weighted) + R = torch.block_diag(*block_R) return R def calc_updown(self, orig_weight): - # this works - # R = self.R - self.R = self.get_weight(self.multiplier()) + oft_blocks = self.oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) - # sending R to device causes major deepfrying i.e. just doesn't work - # R = self.R.to(orig_weight.device, dtype=orig_weight.dtype) + R = self.get_weight(oft_blocks) + self.R = R # if orig_weight.dim() == 4: # weight = torch.einsum("oihw, op -> pihw", orig_weight, R) # else: # weight = torch.einsum("oi, op -> pi", orig_weight, R) - updown = orig_weight @ self.R + updown = orig_weight @ R output_shape = self.oft_blocks.shape - ## this works - # updown = orig_weight @ R - # output_shape = [orig_weight.size(0), R.size(1)] - return self.finalize_updown(updown, orig_weight, output_shape) def forward(self, x, y=None): x = self.org_forward(x) if self.multiplier() == 0.0: return x + + # calculating R here is excruciatingly slow #R = self.get_weight().to(x.device, dtype=x.dtype) R = self.R.to(x.device, dtype=x.dtype) + if x.dim() == 4: x = x.permute(0, 2, 3, 1) x = torch.matmul(x, R) From d10c4db57ed08234a7aed5f530f269ff78544ab0 Mon Sep 17 00:00:00 2001 From: v0xie <28695009+v0xie@users.noreply.github.com> Date: Thu, 19 Oct 2023 12:52:14 -0700 Subject: [PATCH 011/139] style: formatting --- extensions-builtin/Lora/network_oft.py | 4 +-- extensions-builtin/Lora/networks.py | 35 -------------------------- 2 files changed, 2 insertions(+), 37 deletions(-) diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py index 2af1bc4cf..0a87958e2 100644 --- a/extensions-builtin/Lora/network_oft.py +++ b/extensions-builtin/Lora/network_oft.py @@ -37,7 +37,7 @@ class NetworkModuleOFT(network.NetworkModule): def apply_to(self): self.org_forward = self.org_module[0].forward self.org_module[0].forward = self.forward - + def get_weight(self, oft_blocks, multiplier=None): block_Q = oft_blocks - oft_blocks.transpose(1, 2) norm_Q = torch.norm(block_Q.flatten()) @@ -66,7 +66,7 @@ class NetworkModuleOFT(network.NetworkModule): output_shape = self.oft_blocks.shape return self.finalize_updown(updown, orig_weight, output_shape) - + def forward(self, x, y=None): x = self.org_forward(x) if self.multiplier() == 0.0: diff --git a/extensions-builtin/Lora/networks.py b/extensions-builtin/Lora/networks.py index e5e73450b..78a97033d 100644 --- a/extensions-builtin/Lora/networks.py +++ b/extensions-builtin/Lora/networks.py @@ -169,10 +169,6 @@ def load_network(name, network_on_disk): else: emb_dict[vec_name] = weight bundle_embeddings[emb_name] = emb_dict - - #if key_network_without_network_parts == "oft_unet": - # print(key_network_without_network_parts) - # pass key = convert_diffusers_name_to_compvis(key_network_without_network_parts, is_sd2) sd_module = shared.sd_model.network_layer_mapping.get(key, None) @@ -196,31 +192,8 @@ def load_network(name, network_on_disk): sd_module = shared.sd_model.network_layer_mapping.get(key, None) elif sd_module is None and "oft_unet" in key_network_without_network_parts: - # UNET_TARGET_REPLACE_MODULE_ALL_LINEAR = ["Transformer2DModel"] - # UNET_TARGET_REPLACE_MODULE_CONV2D_3X3 = ["ResnetBlock2D", "Downsample2D", "Upsample2D"] - UNET_TARGET_REPLACE_MODULE_ATTN_ONLY = ["CrossAttention"] - # TODO: Change matchedm odules based on whether all linear, conv, etc - key = key_network_without_network_parts.replace("oft_unet", "diffusion_model") sd_module = shared.sd_model.network_layer_mapping.get(key, None) - #key_no_suffix = key.rsplit("_to_", 1)[0] - ## Match all modules of class CrossAttention - #replace_module_list = [] - #for module_type in UNET_TARGET_REPLACE_MODULE_ATTN_ONLY: - # replace_module_list += [module for k, module in shared.sd_model.network_layer_mapping.items() if module_type in module.__class__.__name__] - - #matched_module = replace_module_list.get(key_no_suffix, None) - #if key.endswith('to_q'): - # sd_module = matched_module.to_q or None - #if key.endswith('to_k'): - # sd_module = matched_module.to_k or None - #if key.endswith('to_v'): - # sd_module = matched_module.to_v or None - #if key.endswith('to_out_0'): - # sd_module = matched_module.to_out[0] or None - #if key.endswith('to_out_1'): - # sd_module = matched_module.to_out[1] or None - if sd_module is None: keys_failed_to_match[key_network] = key @@ -242,14 +215,6 @@ def load_network(name, network_on_disk): raise AssertionError(f"Could not find a module type (out of {', '.join([x.__class__.__name__ for x in module_types])}) that would accept those keys: {', '.join(weights.w)}") net.modules[key] = net_module - - # replaces forward method of original Linear - # applied_to_count = 0 - #for key, created_module in net.modules.items(): - # if isinstance(created_module, network_oft.NetworkModuleOFT): - # net_module.apply_to() - #applied_to_count += 1 - # print(f'Applied OFT modules: {applied_to_count}') embeddings = {} for emb_name, data in bundle_embeddings.items(): From 0550659ce6e1c37d1ab05cb8a2cb31d499fa552f Mon Sep 17 00:00:00 2001 From: v0xie <28695009+v0xie@users.noreply.github.com> Date: Thu, 19 Oct 2023 13:13:02 -0700 Subject: [PATCH 012/139] style: fix ambiguous variable name --- extensions-builtin/Lora/network_oft.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py index 0a87958e2..4e8382c18 100644 --- a/extensions-builtin/Lora/network_oft.py +++ b/extensions-builtin/Lora/network_oft.py @@ -43,8 +43,8 @@ class NetworkModuleOFT(network.NetworkModule): norm_Q = torch.norm(block_Q.flatten()) new_norm_Q = torch.clamp(norm_Q, max=self.constraint) block_Q = block_Q * ((new_norm_Q + 1e-8) / (norm_Q + 1e-8)) - I = torch.eye(self.block_size, device=self.oft_blocks.device).unsqueeze(0).repeat(self.num_blocks, 1, 1) - block_R = torch.matmul(I + block_Q, (I - block_Q).inverse()) + m_I = torch.eye(self.block_size, device=self.oft_blocks.device).unsqueeze(0).repeat(self.num_blocks, 1, 1) + block_R = torch.matmul(m_I + block_Q, (m_I - block_Q).inverse()) #block_R_weighted = multiplier * block_R + (1 - multiplier) * I #R = torch.block_diag(*block_R_weighted) R = torch.block_diag(*block_R) From 2d8c894b274d60a3e3563a2ace23c4ebcea9e652 Mon Sep 17 00:00:00 2001 From: v0xie <28695009+v0xie@users.noreply.github.com> Date: Sat, 21 Oct 2023 13:43:31 -0700 Subject: [PATCH 013/139] refactor: use forward hook instead of custom forward --- extensions-builtin/Lora/network_oft.py | 33 +++++++++++++++++++------- 1 file changed, 24 insertions(+), 9 deletions(-) diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py index 4e8382c18..8e561ab0b 100644 --- a/extensions-builtin/Lora/network_oft.py +++ b/extensions-builtin/Lora/network_oft.py @@ -36,9 +36,11 @@ class NetworkModuleOFT(network.NetworkModule): # how do we revert this to unload the weights? def apply_to(self): self.org_forward = self.org_module[0].forward - self.org_module[0].forward = self.forward + #self.org_module[0].forward = self.forward + self.org_module[0].register_forward_hook(self.forward_hook) def get_weight(self, oft_blocks, multiplier=None): + self.constraint = self.constraint.to(oft_blocks.device, dtype=oft_blocks.dtype) block_Q = oft_blocks - oft_blocks.transpose(1, 2) norm_Q = torch.norm(block_Q.flatten()) new_norm_Q = torch.clamp(norm_Q, max=self.constraint) @@ -66,14 +68,10 @@ class NetworkModuleOFT(network.NetworkModule): output_shape = self.oft_blocks.shape return self.finalize_updown(updown, orig_weight, output_shape) - - def forward(self, x, y=None): - x = self.org_forward(x) - if self.multiplier() == 0.0: - return x - - # calculating R here is excruciatingly slow - #R = self.get_weight().to(x.device, dtype=x.dtype) + + def forward_hook(self, module, args, output): + #print(f'Forward hook in {self.network_key} called') + x = output R = self.R.to(x.device, dtype=x.dtype) if x.dim() == 4: @@ -83,3 +81,20 @@ class NetworkModuleOFT(network.NetworkModule): else: x = torch.matmul(x, R) return x + + # def forward(self, x, y=None): + # x = self.org_forward(x) + # if self.multiplier() == 0.0: + # return x + + # # calculating R here is excruciatingly slow + # #R = self.get_weight().to(x.device, dtype=x.dtype) + # R = self.R.to(x.device, dtype=x.dtype) + + # if x.dim() == 4: + # x = x.permute(0, 2, 3, 1) + # x = torch.matmul(x, R) + # x = x.permute(0, 3, 1, 2) + # else: + # x = torch.matmul(x, R) + # return x From 768354772853a1d27a9bf7e41bd6a6e4eac7a9c7 Mon Sep 17 00:00:00 2001 From: v0xie <28695009+v0xie@users.noreply.github.com> Date: Sat, 21 Oct 2023 14:42:24 -0700 Subject: [PATCH 014/139] fix: return orig weights during updown, merge weights before forward --- extensions-builtin/Lora/network_oft.py | 94 +++++++++++++++++++------- 1 file changed, 71 insertions(+), 23 deletions(-) diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py index 8e561ab0b..f5f32c238 100644 --- a/extensions-builtin/Lora/network_oft.py +++ b/extensions-builtin/Lora/network_oft.py @@ -1,5 +1,6 @@ import torch import network +from modules import devices class ModuleTypeOFT(network.ModuleType): @@ -29,23 +30,56 @@ class NetworkModuleOFT(network.NetworkModule): self.block_size = self.out_dim // self.num_blocks self.org_module: list[torch.Module] = [self.sd_module] + self.org_weight = self.org_module[0].weight.to(self.org_module[0].weight.device, copy=True) + #self.org_weight = self.org_module[0].weight.to(devices.cpu, copy=True) self.R = self.get_weight(self.oft_blocks) + + self.merged_weight = self.merge_weight() self.apply_to() + self.merged = False + + + def merge_weight(self): + org_sd = self.org_module[0].state_dict() + R = self.R.to(self.org_weight.device, dtype=self.org_weight.dtype) + if self.org_weight.dim() == 4: + weight = torch.einsum("oihw, op -> pihw", self.org_weight, R) + else: + weight = torch.einsum("oi, op -> pi", self.org_weight, R) + org_sd['weight'] = weight + # replace weight + #self.org_module[0].load_state_dict(org_sd) + return weight + pass + + def replace_weight(self, new_weight): + org_sd = self.org_module[0].state_dict() + org_sd['weight'] = new_weight + self.org_module[0].load_state_dict(org_sd) + self.merged = True + + def restore_weight(self): + org_sd = self.org_module[0].state_dict() + org_sd['weight'] = self.org_weight + self.org_module[0].load_state_dict(org_sd) + self.merged = False + # replace forward method of original linear rather than replacing the module # how do we revert this to unload the weights? def apply_to(self): self.org_forward = self.org_module[0].forward #self.org_module[0].forward = self.forward + self.org_module[0].register_forward_pre_hook(self.pre_forward_hook) self.org_module[0].register_forward_hook(self.forward_hook) def get_weight(self, oft_blocks, multiplier=None): - self.constraint = self.constraint.to(oft_blocks.device, dtype=oft_blocks.dtype) + constraint = self.constraint.to(oft_blocks.device, dtype=oft_blocks.dtype) block_Q = oft_blocks - oft_blocks.transpose(1, 2) norm_Q = torch.norm(block_Q.flatten()) - new_norm_Q = torch.clamp(norm_Q, max=self.constraint) + new_norm_Q = torch.clamp(norm_Q, max=constraint) block_Q = block_Q * ((new_norm_Q + 1e-8) / (norm_Q + 1e-8)) - m_I = torch.eye(self.block_size, device=self.oft_blocks.device).unsqueeze(0).repeat(self.num_blocks, 1, 1) + m_I = torch.eye(self.block_size, device=oft_blocks.device).unsqueeze(0).repeat(self.num_blocks, 1, 1) block_R = torch.matmul(m_I + block_Q, (m_I - block_Q).inverse()) #block_R_weighted = multiplier * block_R + (1 - multiplier) * I #R = torch.block_diag(*block_R_weighted) @@ -54,33 +88,47 @@ class NetworkModuleOFT(network.NetworkModule): return R def calc_updown(self, orig_weight): - oft_blocks = self.oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) + #oft_blocks = self.oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) - R = self.get_weight(oft_blocks) - self.R = R + #R = self.R.to(orig_weight.device, dtype=orig_weight.dtype) + ##self.R = R - # if orig_weight.dim() == 4: - # weight = torch.einsum("oihw, op -> pihw", orig_weight, R) - # else: - # weight = torch.einsum("oi, op -> pi", orig_weight, R) + #if orig_weight.dim() == 4: + # weight = torch.einsum("oihw, op -> pihw", orig_weight, R) + #else: + # weight = torch.einsum("oi, op -> pi", orig_weight, R) - updown = orig_weight @ R - output_shape = self.oft_blocks.shape + #updown = orig_weight @ R + #updown = weight + updown = torch.zeros_like(orig_weight, device=orig_weight.device, dtype=orig_weight.dtype) + #updown = orig_weight + output_shape = orig_weight.shape + #orig_weight = self.merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) + #output_shape = self.oft_blocks.shape return self.finalize_updown(updown, orig_weight, output_shape) - def forward_hook(self, module, args, output): - #print(f'Forward hook in {self.network_key} called') - x = output - R = self.R.to(x.device, dtype=x.dtype) + def pre_forward_hook(self, module, input): + if not self.merged: + self.replace_weight(self.merged_weight) - if x.dim() == 4: - x = x.permute(0, 2, 3, 1) - x = torch.matmul(x, R) - x = x.permute(0, 3, 1, 2) - else: - x = torch.matmul(x, R) - return x + + def forward_hook(self, module, args, output): + if self.merged: + pass + #self.restore_weight() + #print(f'Forward hook in {self.network_key} called') + + #x = output + #R = self.R.to(x.device, dtype=x.dtype) + + #if x.dim() == 4: + # x = x.permute(0, 2, 3, 1) + # x = torch.matmul(x, R) + # x = x.permute(0, 3, 1, 2) + #else: + # x = torch.matmul(x, R) + #return x # def forward(self, x, y=None): # x = self.org_forward(x) From fce86ab7d75690785f0f5b496f1b3aee922c0ae3 Mon Sep 17 00:00:00 2001 From: v0xie <28695009+v0xie@users.noreply.github.com> Date: Sat, 21 Oct 2023 16:03:54 -0700 Subject: [PATCH 015/139] fix: support multiplier, no forward pass hook --- extensions-builtin/Lora/network_oft.py | 43 ++++++++++++++++++++------ 1 file changed, 33 insertions(+), 10 deletions(-) diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py index f5f32c238..e0672ba6d 100644 --- a/extensions-builtin/Lora/network_oft.py +++ b/extensions-builtin/Lora/network_oft.py @@ -32,21 +32,27 @@ class NetworkModuleOFT(network.NetworkModule): self.org_module: list[torch.Module] = [self.sd_module] self.org_weight = self.org_module[0].weight.to(self.org_module[0].weight.device, copy=True) #self.org_weight = self.org_module[0].weight.to(devices.cpu, copy=True) - self.R = self.get_weight(self.oft_blocks) + init_multiplier = self.multiplier() * self.calc_scale() + self.last_multiplier = init_multiplier + self.R = self.get_weight(self.oft_blocks, init_multiplier) self.merged_weight = self.merge_weight() self.apply_to() self.merged = False + # weights_backup = getattr(self.org_module[0], 'network_weights_backup', None) + # if weights_backup is None: + # self.org_module[0].network_weights_backup = self.org_weight + def merge_weight(self): - org_sd = self.org_module[0].state_dict() + #org_sd = self.org_module[0].state_dict() R = self.R.to(self.org_weight.device, dtype=self.org_weight.dtype) if self.org_weight.dim() == 4: weight = torch.einsum("oihw, op -> pihw", self.org_weight, R) else: weight = torch.einsum("oi, op -> pi", self.org_weight, R) - org_sd['weight'] = weight + #org_sd['weight'] = weight # replace weight #self.org_module[0].load_state_dict(org_sd) return weight @@ -74,6 +80,7 @@ class NetworkModuleOFT(network.NetworkModule): self.org_module[0].register_forward_hook(self.forward_hook) def get_weight(self, oft_blocks, multiplier=None): + multiplier = multiplier.to(oft_blocks.device, dtype=oft_blocks.dtype) constraint = self.constraint.to(oft_blocks.device, dtype=oft_blocks.dtype) block_Q = oft_blocks - oft_blocks.transpose(1, 2) norm_Q = torch.norm(block_Q.flatten()) @@ -81,9 +88,9 @@ class NetworkModuleOFT(network.NetworkModule): block_Q = block_Q * ((new_norm_Q + 1e-8) / (norm_Q + 1e-8)) m_I = torch.eye(self.block_size, device=oft_blocks.device).unsqueeze(0).repeat(self.num_blocks, 1, 1) block_R = torch.matmul(m_I + block_Q, (m_I - block_Q).inverse()) - #block_R_weighted = multiplier * block_R + (1 - multiplier) * I - #R = torch.block_diag(*block_R_weighted) - R = torch.block_diag(*block_R) + block_R_weighted = multiplier * block_R + (1 - multiplier) * m_I + R = torch.block_diag(*block_R_weighted) + #R = torch.block_diag(*block_R) return R @@ -93,6 +100,8 @@ class NetworkModuleOFT(network.NetworkModule): #R = self.R.to(orig_weight.device, dtype=orig_weight.dtype) ##self.R = R + #R = self.R.to(orig_weight.device, dtype=orig_weight.dtype) + ##self.R = R #if orig_weight.dim() == 4: # weight = torch.einsum("oihw, op -> pihw", orig_weight, R) #else: @@ -103,19 +112,33 @@ class NetworkModuleOFT(network.NetworkModule): updown = torch.zeros_like(orig_weight, device=orig_weight.device, dtype=orig_weight.dtype) #updown = orig_weight output_shape = orig_weight.shape - #orig_weight = self.merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) + orig_weight = self.merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) #output_shape = self.oft_blocks.shape return self.finalize_updown(updown, orig_weight, output_shape) def pre_forward_hook(self, module, input): - if not self.merged: + multiplier = self.multiplier() * self.calc_scale() + if not multiplier==self.last_multiplier or not self.merged: + + #if multiplier != self.last_multiplier or not self.merged: + self.R = self.get_weight(self.oft_blocks, multiplier) + self.last_multiplier = multiplier + self.merged_weight = self.merge_weight() self.replace_weight(self.merged_weight) + #elif not self.merged: + # self.replace_weight(self.merged_weight) def forward_hook(self, module, args, output): - if self.merged: - pass + pass + #output = output * self.multiplier() * self.calc_scale() + #if len(args) > 0: + # y = args[0] + # output = output + y + #return output + #if self.merged: + # pass #self.restore_weight() #print(f'Forward hook in {self.network_key} called') From 76f5abdbdb739133eff2ccefa36eac62bea3fa08 Mon Sep 17 00:00:00 2001 From: v0xie <28695009+v0xie@users.noreply.github.com> Date: Sat, 21 Oct 2023 16:07:45 -0700 Subject: [PATCH 016/139] style: cleanup oft --- extensions-builtin/Lora/network_oft.py | 82 +++----------------------- 1 file changed, 7 insertions(+), 75 deletions(-) diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py index e0672ba6d..e462ccb1b 100644 --- a/extensions-builtin/Lora/network_oft.py +++ b/extensions-builtin/Lora/network_oft.py @@ -1,6 +1,5 @@ import torch import network -from modules import devices class ModuleTypeOFT(network.ModuleType): @@ -31,33 +30,24 @@ class NetworkModuleOFT(network.NetworkModule): self.org_module: list[torch.Module] = [self.sd_module] self.org_weight = self.org_module[0].weight.to(self.org_module[0].weight.device, copy=True) - #self.org_weight = self.org_module[0].weight.to(devices.cpu, copy=True) + init_multiplier = self.multiplier() * self.calc_scale() self.last_multiplier = init_multiplier + self.R = self.get_weight(self.oft_blocks, init_multiplier) self.merged_weight = self.merge_weight() self.apply_to() self.merged = False - # weights_backup = getattr(self.org_module[0], 'network_weights_backup', None) - # if weights_backup is None: - # self.org_module[0].network_weights_backup = self.org_weight - - def merge_weight(self): - #org_sd = self.org_module[0].state_dict() R = self.R.to(self.org_weight.device, dtype=self.org_weight.dtype) if self.org_weight.dim() == 4: weight = torch.einsum("oihw, op -> pihw", self.org_weight, R) else: weight = torch.einsum("oi, op -> pi", self.org_weight, R) - #org_sd['weight'] = weight - # replace weight - #self.org_module[0].load_state_dict(org_sd) return weight - pass - + def replace_weight(self, new_weight): org_sd = self.org_module[0].state_dict() org_sd['weight'] = new_weight @@ -70,9 +60,7 @@ class NetworkModuleOFT(network.NetworkModule): self.org_module[0].load_state_dict(org_sd) self.merged = False - - # replace forward method of original linear rather than replacing the module - # how do we revert this to unload the weights? + # FIXME: hook forward method of original linear, but how do we undo the hook when we are done? def apply_to(self): self.org_forward = self.org_module[0].forward #self.org_module[0].forward = self.forward @@ -90,82 +78,26 @@ class NetworkModuleOFT(network.NetworkModule): block_R = torch.matmul(m_I + block_Q, (m_I - block_Q).inverse()) block_R_weighted = multiplier * block_R + (1 - multiplier) * m_I R = torch.block_diag(*block_R_weighted) - #R = torch.block_diag(*block_R) return R def calc_updown(self, orig_weight): - #oft_blocks = self.oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) - - #R = self.R.to(orig_weight.device, dtype=orig_weight.dtype) - ##self.R = R - - #R = self.R.to(orig_weight.device, dtype=orig_weight.dtype) - ##self.R = R - #if orig_weight.dim() == 4: - # weight = torch.einsum("oihw, op -> pihw", orig_weight, R) - #else: - # weight = torch.einsum("oi, op -> pi", orig_weight, R) - - #updown = orig_weight @ R - #updown = weight updown = torch.zeros_like(orig_weight, device=orig_weight.device, dtype=orig_weight.dtype) - #updown = orig_weight output_shape = orig_weight.shape orig_weight = self.merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) #output_shape = self.oft_blocks.shape return self.finalize_updown(updown, orig_weight, output_shape) - + def pre_forward_hook(self, module, input): multiplier = self.multiplier() * self.calc_scale() - if not multiplier==self.last_multiplier or not self.merged: - #if multiplier != self.last_multiplier or not self.merged: + if not multiplier==self.last_multiplier or not self.merged: self.R = self.get_weight(self.oft_blocks, multiplier) self.last_multiplier = multiplier self.merged_weight = self.merge_weight() self.replace_weight(self.merged_weight) - #elif not self.merged: - # self.replace_weight(self.merged_weight) - + def forward_hook(self, module, args, output): pass - #output = output * self.multiplier() * self.calc_scale() - #if len(args) > 0: - # y = args[0] - # output = output + y - #return output - #if self.merged: - # pass - #self.restore_weight() - #print(f'Forward hook in {self.network_key} called') - - #x = output - #R = self.R.to(x.device, dtype=x.dtype) - - #if x.dim() == 4: - # x = x.permute(0, 2, 3, 1) - # x = torch.matmul(x, R) - # x = x.permute(0, 3, 1, 2) - #else: - # x = torch.matmul(x, R) - #return x - - # def forward(self, x, y=None): - # x = self.org_forward(x) - # if self.multiplier() == 0.0: - # return x - - # # calculating R here is excruciatingly slow - # #R = self.get_weight().to(x.device, dtype=x.dtype) - # R = self.R.to(x.device, dtype=x.dtype) - - # if x.dim() == 4: - # x = x.permute(0, 2, 3, 1) - # x = torch.matmul(x, R) - # x = x.permute(0, 3, 1, 2) - # else: - # x = torch.matmul(x, R) - # return x From de8ee92ed88b855098e273f576a27f4789f0693d Mon Sep 17 00:00:00 2001 From: v0xie <28695009+v0xie@users.noreply.github.com> Date: Sat, 21 Oct 2023 17:37:17 -0700 Subject: [PATCH 017/139] fix: use merge_weight to cache value --- extensions-builtin/Lora/network_oft.py | 57 ++++++++++++++++++-------- 1 file changed, 40 insertions(+), 17 deletions(-) diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py index e462ccb1b..ebe6740c5 100644 --- a/extensions-builtin/Lora/network_oft.py +++ b/extensions-builtin/Lora/network_oft.py @@ -29,23 +29,27 @@ class NetworkModuleOFT(network.NetworkModule): self.block_size = self.out_dim // self.num_blocks self.org_module: list[torch.Module] = [self.sd_module] - self.org_weight = self.org_module[0].weight.to(self.org_module[0].weight.device, copy=True) + #self.org_weight = self.org_module[0].weight.to(self.org_module[0].weight.device, copy=True) init_multiplier = self.multiplier() * self.calc_scale() self.last_multiplier = init_multiplier self.R = self.get_weight(self.oft_blocks, init_multiplier) + self.hooks = [] self.merged_weight = self.merge_weight() - self.apply_to() + + #self.apply_to() + self.applied = False self.merged = False def merge_weight(self): - R = self.R.to(self.org_weight.device, dtype=self.org_weight.dtype) - if self.org_weight.dim() == 4: - weight = torch.einsum("oihw, op -> pihw", self.org_weight, R) + org_weight = self.org_module[0].weight + R = self.R.to(org_weight.device, dtype=org_weight.dtype) + if org_weight.dim() == 4: + weight = torch.einsum("oihw, op -> pihw", org_weight, R) else: - weight = torch.einsum("oi, op -> pi", self.org_weight, R) + weight = torch.einsum("oi, op -> pi", org_weight, R) return weight def replace_weight(self, new_weight): @@ -55,17 +59,29 @@ class NetworkModuleOFT(network.NetworkModule): self.merged = True def restore_weight(self): - org_sd = self.org_module[0].state_dict() - org_sd['weight'] = self.org_weight - self.org_module[0].load_state_dict(org_sd) - self.merged = False + pass + #org_sd = self.org_module[0].state_dict() + #org_sd['weight'] = self.org_weight + #self.org_module[0].load_state_dict(org_sd) + #self.merged = False # FIXME: hook forward method of original linear, but how do we undo the hook when we are done? def apply_to(self): - self.org_forward = self.org_module[0].forward - #self.org_module[0].forward = self.forward - self.org_module[0].register_forward_pre_hook(self.pre_forward_hook) - self.org_module[0].register_forward_hook(self.forward_hook) + if not self.applied: + self.org_forward = self.org_module[0].forward + #self.org_module[0].forward = self.forward + prehook = self.org_module[0].register_forward_pre_hook(self.pre_forward_hook) + hook = self.org_module[0].register_forward_hook(self.forward_hook) + self.hooks.append(prehook) + self.hooks.append(hook) + self.applied = True + + def remove_from(self): + if self.applied: + for hook in self.hooks: + hook.remove() + self.hooks = [] + self.applied = False def get_weight(self, oft_blocks, multiplier=None): multiplier = multiplier.to(oft_blocks.device, dtype=oft_blocks.dtype) @@ -82,14 +98,22 @@ class NetworkModuleOFT(network.NetworkModule): return R def calc_updown(self, orig_weight): + if not self.applied: + self.apply_to() + + self.merged_weight = self.merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) + updown = torch.zeros_like(orig_weight, device=orig_weight.device, dtype=orig_weight.dtype) output_shape = orig_weight.shape - orig_weight = self.merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) + orig_weight = self.merged_weight #output_shape = self.oft_blocks.shape return self.finalize_updown(updown, orig_weight, output_shape) def pre_forward_hook(self, module, input): + #if not self.applied: + # self.apply_to() + multiplier = self.multiplier() * self.calc_scale() if not multiplier==self.last_multiplier or not self.merged: @@ -98,6 +122,5 @@ class NetworkModuleOFT(network.NetworkModule): self.merged_weight = self.merge_weight() self.replace_weight(self.merged_weight) - def forward_hook(self, module, args, output): - pass + pass \ No newline at end of file From 4a50c9638c3eac860fb05ae603cd61aabf4cd1a9 Mon Sep 17 00:00:00 2001 From: v0xie <28695009+v0xie@users.noreply.github.com> Date: Sun, 22 Oct 2023 08:54:24 -0700 Subject: [PATCH 018/139] refactor: remove used OFT functions --- extensions-builtin/Lora/network_oft.py | 82 ++++---------------------- 1 file changed, 10 insertions(+), 72 deletions(-) diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py index ebe6740c5..3034a407e 100644 --- a/extensions-builtin/Lora/network_oft.py +++ b/extensions-builtin/Lora/network_oft.py @@ -29,98 +29,36 @@ class NetworkModuleOFT(network.NetworkModule): self.block_size = self.out_dim // self.num_blocks self.org_module: list[torch.Module] = [self.sd_module] - #self.org_weight = self.org_module[0].weight.to(self.org_module[0].weight.device, copy=True) - init_multiplier = self.multiplier() * self.calc_scale() - self.last_multiplier = init_multiplier - - self.R = self.get_weight(self.oft_blocks, init_multiplier) - - self.hooks = [] - self.merged_weight = self.merge_weight() - - #self.apply_to() - self.applied = False - self.merged = False - - def merge_weight(self): - org_weight = self.org_module[0].weight - R = self.R.to(org_weight.device, dtype=org_weight.dtype) + def merge_weight(self, R_weight, org_weight): + R_weight = R_weight.to(org_weight.device, dtype=org_weight.dtype) if org_weight.dim() == 4: - weight = torch.einsum("oihw, op -> pihw", org_weight, R) + weight = torch.einsum("oihw, op -> pihw", org_weight, R_weight) else: - weight = torch.einsum("oi, op -> pi", org_weight, R) + weight = torch.einsum("oi, op -> pi", org_weight, R_weight) return weight - def replace_weight(self, new_weight): - org_sd = self.org_module[0].state_dict() - org_sd['weight'] = new_weight - self.org_module[0].load_state_dict(org_sd) - self.merged = True - - def restore_weight(self): - pass - #org_sd = self.org_module[0].state_dict() - #org_sd['weight'] = self.org_weight - #self.org_module[0].load_state_dict(org_sd) - #self.merged = False - - # FIXME: hook forward method of original linear, but how do we undo the hook when we are done? - def apply_to(self): - if not self.applied: - self.org_forward = self.org_module[0].forward - #self.org_module[0].forward = self.forward - prehook = self.org_module[0].register_forward_pre_hook(self.pre_forward_hook) - hook = self.org_module[0].register_forward_hook(self.forward_hook) - self.hooks.append(prehook) - self.hooks.append(hook) - self.applied = True - - def remove_from(self): - if self.applied: - for hook in self.hooks: - hook.remove() - self.hooks = [] - self.applied = False - def get_weight(self, oft_blocks, multiplier=None): - multiplier = multiplier.to(oft_blocks.device, dtype=oft_blocks.dtype) constraint = self.constraint.to(oft_blocks.device, dtype=oft_blocks.dtype) + block_Q = oft_blocks - oft_blocks.transpose(1, 2) norm_Q = torch.norm(block_Q.flatten()) new_norm_Q = torch.clamp(norm_Q, max=constraint) block_Q = block_Q * ((new_norm_Q + 1e-8) / (norm_Q + 1e-8)) m_I = torch.eye(self.block_size, device=oft_blocks.device).unsqueeze(0).repeat(self.num_blocks, 1, 1) block_R = torch.matmul(m_I + block_Q, (m_I - block_Q).inverse()) + block_R_weighted = multiplier * block_R + (1 - multiplier) * m_I R = torch.block_diag(*block_R_weighted) return R def calc_updown(self, orig_weight): - if not self.applied: - self.apply_to() + R = self.get_weight(self.oft_blocks, self.multiplier()) + merged_weight = self.merge_weight(R, orig_weight) - self.merged_weight = self.merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) - - updown = torch.zeros_like(orig_weight, device=orig_weight.device, dtype=orig_weight.dtype) + updown = merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight output_shape = orig_weight.shape - orig_weight = self.merged_weight - #output_shape = self.oft_blocks.shape + orig_weight = orig_weight return self.finalize_updown(updown, orig_weight, output_shape) - - def pre_forward_hook(self, module, input): - #if not self.applied: - # self.apply_to() - - multiplier = self.multiplier() * self.calc_scale() - - if not multiplier==self.last_multiplier or not self.merged: - self.R = self.get_weight(self.oft_blocks, multiplier) - self.last_multiplier = multiplier - self.merged_weight = self.merge_weight() - self.replace_weight(self.merged_weight) - - def forward_hook(self, module, args, output): - pass \ No newline at end of file From 3b8515d2c9abad7f0ccaac0215803716e861ee0e Mon Sep 17 00:00:00 2001 From: v0xie <28695009+v0xie@users.noreply.github.com> Date: Sun, 22 Oct 2023 09:27:48 -0700 Subject: [PATCH 019/139] fix: multiplier applied twice in finalize_updown --- extensions-builtin/Lora/network_oft.py | 23 ++++++++++++++++++++++- 1 file changed, 22 insertions(+), 1 deletion(-) diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py index 3034a407e..efbdd296a 100644 --- a/extensions-builtin/Lora/network_oft.py +++ b/extensions-builtin/Lora/network_oft.py @@ -54,7 +54,8 @@ class NetworkModuleOFT(network.NetworkModule): return R def calc_updown(self, orig_weight): - R = self.get_weight(self.oft_blocks, self.multiplier()) + multiplier = self.multiplier() * self.calc_scale() + R = self.get_weight(self.oft_blocks, multiplier) merged_weight = self.merge_weight(R, orig_weight) updown = merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight @@ -62,3 +63,23 @@ class NetworkModuleOFT(network.NetworkModule): orig_weight = orig_weight return self.finalize_updown(updown, orig_weight, output_shape) + + # override to remove the multiplier/scale factor; it's already multiplied in get_weight + def finalize_updown(self, updown, orig_weight, output_shape, ex_bias=None): + #return super().finalize_updown(updown, orig_weight, output_shape, ex_bias) + + if self.bias is not None: + updown = updown.reshape(self.bias.shape) + updown += self.bias.to(orig_weight.device, dtype=orig_weight.dtype) + updown = updown.reshape(output_shape) + + if len(output_shape) == 4: + updown = updown.reshape(output_shape) + + if orig_weight.size().numel() == updown.size().numel(): + updown = updown.reshape(orig_weight.shape) + + if ex_bias is not None: + ex_bias = ex_bias * self.multiplier() + + return updown, ex_bias From 6523edb8a45d4e09f11f3b4e1d133afa6fb65e53 Mon Sep 17 00:00:00 2001 From: v0xie <28695009+v0xie@users.noreply.github.com> Date: Sun, 22 Oct 2023 09:31:15 -0700 Subject: [PATCH 020/139] style: conform style --- extensions-builtin/Lora/network_oft.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py index efbdd296a..e43c9a1df 100644 --- a/extensions-builtin/Lora/network_oft.py +++ b/extensions-builtin/Lora/network_oft.py @@ -63,7 +63,7 @@ class NetworkModuleOFT(network.NetworkModule): orig_weight = orig_weight return self.finalize_updown(updown, orig_weight, output_shape) - + # override to remove the multiplier/scale factor; it's already multiplied in get_weight def finalize_updown(self, updown, orig_weight, output_shape, ex_bias=None): #return super().finalize_updown(updown, orig_weight, output_shape, ex_bias) From a2fad6ee055f3f4e98e46b6c2d912776fe608214 Mon Sep 17 00:00:00 2001 From: v0xie <28695009+v0xie@users.noreply.github.com> Date: Wed, 1 Nov 2023 22:34:27 -0700 Subject: [PATCH 021/139] test implementation based on kohaku diag-oft implementation --- extensions-builtin/Lora/network_oft.py | 57 +++++++++++++++++--------- 1 file changed, 37 insertions(+), 20 deletions(-) diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py index e43c9a1df..ff61b3699 100644 --- a/extensions-builtin/Lora/network_oft.py +++ b/extensions-builtin/Lora/network_oft.py @@ -1,5 +1,6 @@ import torch import network +from einops import rearrange class ModuleTypeOFT(network.ModuleType): @@ -30,35 +31,51 @@ class NetworkModuleOFT(network.NetworkModule): self.org_module: list[torch.Module] = [self.sd_module] - def merge_weight(self, R_weight, org_weight): - R_weight = R_weight.to(org_weight.device, dtype=org_weight.dtype) - if org_weight.dim() == 4: - weight = torch.einsum("oihw, op -> pihw", org_weight, R_weight) - else: - weight = torch.einsum("oi, op -> pi", org_weight, R_weight) - return weight + # def merge_weight(self, R_weight, org_weight): + # R_weight = R_weight.to(org_weight.device, dtype=org_weight.dtype) + # if org_weight.dim() == 4: + # weight = torch.einsum("oihw, op -> pihw", org_weight, R_weight) + # else: + # weight = torch.einsum("oi, op -> pi", org_weight, R_weight) + # weight = torch.einsum( + # "k n m, k n ... -> k m ...", + # self.oft_diag * scale + torch.eye(self.block_size, device=device), + # org_weight + # ) + # return weight def get_weight(self, oft_blocks, multiplier=None): - constraint = self.constraint.to(oft_blocks.device, dtype=oft_blocks.dtype) + # constraint = self.constraint.to(oft_blocks.device, dtype=oft_blocks.dtype) - block_Q = oft_blocks - oft_blocks.transpose(1, 2) - norm_Q = torch.norm(block_Q.flatten()) - new_norm_Q = torch.clamp(norm_Q, max=constraint) - block_Q = block_Q * ((new_norm_Q + 1e-8) / (norm_Q + 1e-8)) - m_I = torch.eye(self.block_size, device=oft_blocks.device).unsqueeze(0).repeat(self.num_blocks, 1, 1) - block_R = torch.matmul(m_I + block_Q, (m_I - block_Q).inverse()) + # block_Q = oft_blocks - oft_blocks.transpose(1, 2) + # norm_Q = torch.norm(block_Q.flatten()) + # new_norm_Q = torch.clamp(norm_Q, max=constraint) + # block_Q = block_Q * ((new_norm_Q + 1e-8) / (norm_Q + 1e-8)) + # m_I = torch.eye(self.block_size, device=oft_blocks.device).unsqueeze(0).repeat(self.num_blocks, 1, 1) + # block_R = torch.matmul(m_I + block_Q, (m_I - block_Q).inverse()) - block_R_weighted = multiplier * block_R + (1 - multiplier) * m_I - R = torch.block_diag(*block_R_weighted) + # block_R_weighted = multiplier * block_R + (1 - multiplier) * m_I + # R = torch.block_diag(*block_R_weighted) + #return R + return self.oft_blocks - return R def calc_updown(self, orig_weight): multiplier = self.multiplier() * self.calc_scale() - R = self.get_weight(self.oft_blocks, multiplier) - merged_weight = self.merge_weight(R, orig_weight) + #R = self.get_weight(self.oft_blocks, multiplier) + R = self.oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) + #merged_weight = self.merge_weight(R, orig_weight) - updown = merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight + orig_weight = rearrange(orig_weight, '(k n) ... -> k n ...', k=self.num_blocks, n=self.block_size) + weight = torch.einsum( + 'k n m, k n ... -> k m ...', + R * multiplier + torch.eye(self.block_size, device=orig_weight.device), + orig_weight + ) + weight = rearrange(weight, 'k m ... -> (k m) ...') + + #updown = merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight + updown = weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight output_shape = orig_weight.shape orig_weight = orig_weight From 65ccd6305fcf72347d5ed68f03095dced865ef6e Mon Sep 17 00:00:00 2001 From: v0xie <28695009+v0xie@users.noreply.github.com> Date: Thu, 2 Nov 2023 00:11:32 -0700 Subject: [PATCH 022/139] detect diag_oft type --- extensions-builtin/Lora/networks.py | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/extensions-builtin/Lora/networks.py b/extensions-builtin/Lora/networks.py index 78a97033d..7f814706a 100644 --- a/extensions-builtin/Lora/networks.py +++ b/extensions-builtin/Lora/networks.py @@ -191,10 +191,17 @@ def load_network(name, network_on_disk): key = key_network_without_network_parts.replace("lora_te1_text_model", "transformer_text_model") sd_module = shared.sd_model.network_layer_mapping.get(key, None) + # kohya_ss OFT module elif sd_module is None and "oft_unet" in key_network_without_network_parts: key = key_network_without_network_parts.replace("oft_unet", "diffusion_model") sd_module = shared.sd_model.network_layer_mapping.get(key, None) + # KohakuBlueLeaf OFT module + if sd_module is None and "oft_diag" in key: + key = key_network_without_network_parts.replace("lora_unet", "diffusion_model") + key = key_network_without_network_parts.replace("lora_te1_text_model", "0_transformer_text_model") + sd_module = shared.sd_model.network_layer_mapping.get(key, None) + if sd_module is None: keys_failed_to_match[key_network] = key continue From d727ddfccdc6d474767be9dc3bf504150e81a8a5 Mon Sep 17 00:00:00 2001 From: v0xie <28695009+v0xie@users.noreply.github.com> Date: Thu, 2 Nov 2023 00:13:11 -0700 Subject: [PATCH 023/139] no idea what i'm doing, trying to support both type of OFT, kblueleaf diag_oft has MultiheadAttn which kohya's doesn't?, attempt create new module based off network_lora.py, errors about tensor dim mismatch --- extensions-builtin/Lora/network_oft.py | 186 +++++++++++++++++++------ 1 file changed, 142 insertions(+), 44 deletions(-) diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py index ff61b3699..e102eafc1 100644 --- a/extensions-builtin/Lora/network_oft.py +++ b/extensions-builtin/Lora/network_oft.py @@ -1,11 +1,12 @@ import torch import network from einops import rearrange +from modules import devices class ModuleTypeOFT(network.ModuleType): def create_module(self, net: network.Network, weights: network.NetworkWeights): - if all(x in weights.w for x in ["oft_blocks"]): + if all(x in weights.w for x in ["oft_blocks"]) or all(x in weights.w for x in ["oft_diag"]): return NetworkModuleOFT(net, weights) return None @@ -16,66 +17,117 @@ class NetworkModuleOFT(network.NetworkModule): super().__init__(net, weights) - self.oft_blocks = weights.w["oft_blocks"] - self.alpha = weights.w["alpha"] - self.dim = self.oft_blocks.shape[0] - self.num_blocks = self.dim + self.lin_module = None + # kohya-ss + if "oft_blocks" in weights.w.keys(): + self.is_kohya = True + self.oft_blocks = weights.w["oft_blocks"] + self.alpha = weights.w["alpha"] + self.dim = self.oft_blocks.shape[0] + elif "oft_diag" in weights.w.keys(): + self.is_kohya = False + self.oft_blocks = weights.w["oft_diag"] + # alpha is rank if alpha is 0 or None + if self.alpha is None: + pass + self.dim = self.oft_blocks.shape[0] # FIXME: almost certainly incorrect, assumes tensor is shape [*, m, n] + else: + raise ValueError("oft_blocks or oft_diag must be in weights dict") - if "Linear" in self.sd_module.__class__.__name__: + is_linear = type(self.sd_module) in [torch.nn.Linear, torch.nn.modules.linear.NonDynamicallyQuantizableLinear] + is_conv = type(self.sd_module) in [torch.nn.Conv2d] + is_other_linear = type(self.sd_module) in [ torch.nn.MultiheadAttention] + #if "Linear" in self.sd_module.__class__.__name__ or is_linear: + if is_linear: self.out_dim = self.sd_module.out_features - elif "Conv" in self.sd_module.__class__.__name__: + #elif hasattr(self.sd_module, "embed_dim"): + # self.out_dim = self.sd_module.embed_dim + #else: + # raise ValueError("Linear sd_module must have out_features or embed_dim") + elif is_other_linear: + self.out_dim = self.sd_module.embed_dim + elif is_conv: self.out_dim = self.sd_module.out_channels + else: + raise ValueError("sd_module must be Linear or Conv") - self.constraint = self.alpha * self.out_dim - self.block_size = self.out_dim // self.num_blocks + + if self.is_kohya: + self.num_blocks = self.dim + self.block_size = self.out_dim // self.num_blocks + self.constraint = self.alpha * self.out_dim + #elif is_linear or is_conv: + else: + self.num_blocks, self.block_size = factorization(self.out_dim, self.dim) + self.constraint = None self.org_module: list[torch.Module] = [self.sd_module] - # def merge_weight(self, R_weight, org_weight): - # R_weight = R_weight.to(org_weight.device, dtype=org_weight.dtype) - # if org_weight.dim() == 4: - # weight = torch.einsum("oihw, op -> pihw", org_weight, R_weight) - # else: - # weight = torch.einsum("oi, op -> pi", org_weight, R_weight) - # weight = torch.einsum( - # "k n m, k n ... -> k m ...", - # self.oft_diag * scale + torch.eye(self.block_size, device=device), - # org_weight - # ) - # return weight + # if is_other_linear: + # weight = self.oft_blocks.reshape(self.oft_blocks.shape[0], -1) + # module = torch.nn.Linear(weight.shape[1], weight.shape[0], bias=False) + # with torch.no_grad(): + # if weight.shape != module.weight.shape: + # weight = weight.reshape(module.weight.shape) + # module.weight.copy_(weight) + # module.to(device=devices.cpu, dtype=devices.dtype) + # module.weight.requires_grad_(False) + # self.lin_module = module + #return module + + def merge_weight(self, R_weight, org_weight): + R_weight = R_weight.to(org_weight.device, dtype=org_weight.dtype) + if org_weight.dim() == 4: + weight = torch.einsum("oihw, op -> pihw", org_weight, R_weight) + else: + weight = torch.einsum("oi, op -> pi", org_weight, R_weight) + #weight = torch.einsum( + # "k n m, k n ... -> k m ...", + # self.oft_diag * scale + torch.eye(self.block_size, device=device), + # org_weight + #) + return weight def get_weight(self, oft_blocks, multiplier=None): - # constraint = self.constraint.to(oft_blocks.device, dtype=oft_blocks.dtype) + if self.constraint is not None: + constraint = self.constraint.to(oft_blocks.device, dtype=oft_blocks.dtype) - # block_Q = oft_blocks - oft_blocks.transpose(1, 2) - # norm_Q = torch.norm(block_Q.flatten()) - # new_norm_Q = torch.clamp(norm_Q, max=constraint) - # block_Q = block_Q * ((new_norm_Q + 1e-8) / (norm_Q + 1e-8)) - # m_I = torch.eye(self.block_size, device=oft_blocks.device).unsqueeze(0).repeat(self.num_blocks, 1, 1) - # block_R = torch.matmul(m_I + block_Q, (m_I - block_Q).inverse()) + block_Q = oft_blocks - oft_blocks.transpose(1, 2) + norm_Q = torch.norm(block_Q.flatten()) + if self.constraint is not None: + new_norm_Q = torch.clamp(norm_Q, max=constraint) + else: + new_norm_Q = norm_Q + block_Q = block_Q * ((new_norm_Q + 1e-8) / (norm_Q + 1e-8)) + m_I = torch.eye(self.block_size, device=oft_blocks.device).unsqueeze(0).repeat(self.num_blocks, 1, 1) + block_R = torch.matmul(m_I + block_Q, (m_I - block_Q).inverse()) - # block_R_weighted = multiplier * block_R + (1 - multiplier) * m_I - # R = torch.block_diag(*block_R_weighted) - #return R - return self.oft_blocks + block_R_weighted = multiplier * block_R + (1 - multiplier) * m_I + R = torch.block_diag(*block_R_weighted) + return R + #return self.oft_blocks def calc_updown(self, orig_weight): multiplier = self.multiplier() * self.calc_scale() - #R = self.get_weight(self.oft_blocks, multiplier) - R = self.oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) - #merged_weight = self.merge_weight(R, orig_weight) + R = self.get_weight(self.oft_blocks, multiplier) + #R = self.oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) + merged_weight = self.merge_weight(R, orig_weight) - orig_weight = rearrange(orig_weight, '(k n) ... -> k n ...', k=self.num_blocks, n=self.block_size) - weight = torch.einsum( - 'k n m, k n ... -> k m ...', - R * multiplier + torch.eye(self.block_size, device=orig_weight.device), - orig_weight - ) - weight = rearrange(weight, 'k m ... -> (k m) ...') + #if self.lin_module is not None: + # R = self.lin_module.weight.to(orig_weight.device, dtype=orig_weight.dtype) + # weight = torch.mul(torch.mul(R, multiplier), orig_weight) + #else: + # orig_weight = rearrange(orig_weight, '(k n) ... -> k n ...', k=self.num_blocks, n=self.block_size) + # weight = torch.einsum( + # 'k n m, k n ... -> k m ...', + # R * multiplier + torch.eye(self.block_size, device=orig_weight.device), + # orig_weight + # ) + # weight = rearrange(weight, 'k m ... -> (k m) ...') - #updown = merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight - updown = weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight + updown = merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight + #updown = weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight output_shape = orig_weight.shape orig_weight = orig_weight @@ -100,3 +152,49 @@ class NetworkModuleOFT(network.NetworkModule): ex_bias = ex_bias * self.multiplier() return updown, ex_bias + +# copied from https://github.com/KohakuBlueleaf/LyCORIS/blob/dev/lycoris/modules/lokr.py +def factorization(dimension: int, factor:int=-1) -> tuple[int, int]: + ''' + return a tuple of two value of input dimension decomposed by the number closest to factor + second value is higher or equal than first value. + + In LoRA with Kroneckor Product, first value is a value for weight scale. + secon value is a value for weight. + + Becuase of non-commutative property, A⊗B ≠ B⊗A. Meaning of two matrices is slightly different. + + examples) + factor + -1 2 4 8 16 ... + 127 -> 1, 127 127 -> 1, 127 127 -> 1, 127 127 -> 1, 127 127 -> 1, 127 + 128 -> 8, 16 128 -> 2, 64 128 -> 4, 32 128 -> 8, 16 128 -> 8, 16 + 250 -> 10, 25 250 -> 2, 125 250 -> 2, 125 250 -> 5, 50 250 -> 10, 25 + 360 -> 8, 45 360 -> 2, 180 360 -> 4, 90 360 -> 8, 45 360 -> 12, 30 + 512 -> 16, 32 512 -> 2, 256 512 -> 4, 128 512 -> 8, 64 512 -> 16, 32 + 1024 -> 32, 32 1024 -> 2, 512 1024 -> 4, 256 1024 -> 8, 128 1024 -> 16, 64 + ''' + + if factor > 0 and (dimension % factor) == 0: + m = factor + n = dimension // factor + if m > n: + n, m = m, n + return m, n + if factor < 0: + factor = dimension + m, n = 1, dimension + length = m + n + while m length or new_m>factor: + break + else: + m, n = new_m, new_n + if m > n: + n, m = m, n + return m, n + From 759515316e8ec536f34fad616e8c6a33674a164b Mon Sep 17 00:00:00 2001 From: Emily Zeng Date: Thu, 2 Nov 2023 21:54:48 -0400 Subject: [PATCH 024/139] added accordion settings options --- modules/shared_options.py | 2 + modules/ui.py | 480 +++++++++++++++++++------------------- 2 files changed, 243 insertions(+), 239 deletions(-) diff --git a/modules/shared_options.py b/modules/shared_options.py index 0a82216ff..5b07dd041 100644 --- a/modules/shared_options.py +++ b/modules/shared_options.py @@ -270,6 +270,8 @@ options_templates.update(options_section(('ui', "User interface"), { "hires_fix_show_sampler": OptionInfo(False, "Hires fix: show hires checkpoint and sampler selection").needs_reload_ui(), "hires_fix_show_prompts": OptionInfo(False, "Hires fix: show hires prompt and negative prompt").needs_reload_ui(), "disable_token_counters": OptionInfo(False, "Disable prompt token counters").needs_reload_ui(), + "txt2img_settings_accordion": OptionInfo(False, "Settings in txt2img hidden under Accordion").needs_reload_ui(), + "img2img_settings_accordion": OptionInfo(False, "Settings in img2img hidden under Accordion").needs_reload_ui(), })) diff --git a/modules/ui.py b/modules/ui.py index bcf391997..d05b9f55d 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -344,85 +344,86 @@ def create_ui(): extra_tabs.__enter__() with gr.Tab("Generation", id="txt2img_generation") as txt2img_generation_tab, ResizeHandleRow(equal_height=False): - with gr.Column(variant='compact', elem_id="txt2img_settings"): - scripts.scripts_txt2img.prepare_ui() + with gr.Accordion("Open for Settings", open=False) if shared.opts.img2img_settings_accordion else gr.Group(): + with gr.Column(variant='compact', elem_id="txt2img_settings"): + scripts.scripts_txt2img.prepare_ui() - for category in ordered_ui_categories(): - if category == "sampler": - steps, sampler_name = create_sampler_and_steps_selection(sd_samplers.visible_sampler_names(), "txt2img") + for category in ordered_ui_categories(): + if category == "sampler": + steps, sampler_name = create_sampler_and_steps_selection(sd_samplers.visible_sampler_names(), "txt2img") - elif category == "dimensions": - with FormRow(): - with gr.Column(elem_id="txt2img_column_size", scale=4): - width = gr.Slider(minimum=64, maximum=2048, step=8, label="Width", value=512, elem_id="txt2img_width") - height = gr.Slider(minimum=64, maximum=2048, step=8, label="Height", value=512, elem_id="txt2img_height") + elif category == "dimensions": + with FormRow(): + with gr.Column(elem_id="txt2img_column_size", scale=4): + width = gr.Slider(minimum=64, maximum=2048, step=8, label="Width", value=512, elem_id="txt2img_width") + height = gr.Slider(minimum=64, maximum=2048, step=8, label="Height", value=512, elem_id="txt2img_height") - with gr.Column(elem_id="txt2img_dimensions_row", scale=1, elem_classes="dimensions-tools"): - res_switch_btn = ToolButton(value=switch_values_symbol, elem_id="txt2img_res_switch_btn", tooltip="Switch width/height") + with gr.Column(elem_id="txt2img_dimensions_row", scale=1, elem_classes="dimensions-tools"): + res_switch_btn = ToolButton(value=switch_values_symbol, elem_id="txt2img_res_switch_btn", tooltip="Switch width/height") - if opts.dimensions_and_batch_together: - with gr.Column(elem_id="txt2img_column_batch"): + if opts.dimensions_and_batch_together: + with gr.Column(elem_id="txt2img_column_batch"): + batch_count = gr.Slider(minimum=1, step=1, label='Batch count', value=1, elem_id="txt2img_batch_count") + batch_size = gr.Slider(minimum=1, maximum=8, step=1, label='Batch size', value=1, elem_id="txt2img_batch_size") + + elif category == "cfg": + with gr.Row(): + cfg_scale = gr.Slider(minimum=1.0, maximum=30.0, step=0.5, label='CFG Scale', value=7.0, elem_id="txt2img_cfg_scale") + + elif category == "checkboxes": + with FormRow(elem_classes="checkboxes-row", variant="compact"): + pass + + elif category == "accordions": + with gr.Row(elem_id="txt2img_accordions", elem_classes="accordions"): + with InputAccordion(False, label="Hires. fix", elem_id="txt2img_hr") as enable_hr: + with enable_hr.extra(): + hr_final_resolution = FormHTML(value="", elem_id="txtimg_hr_finalres", label="Upscaled resolution", interactive=False, min_width=0) + + with FormRow(elem_id="txt2img_hires_fix_row1", variant="compact"): + hr_upscaler = gr.Dropdown(label="Upscaler", elem_id="txt2img_hr_upscaler", choices=[*shared.latent_upscale_modes, *[x.name for x in shared.sd_upscalers]], value=shared.latent_upscale_default_mode) + hr_second_pass_steps = gr.Slider(minimum=0, maximum=150, step=1, label='Hires steps', value=0, elem_id="txt2img_hires_steps") + denoising_strength = gr.Slider(minimum=0.0, maximum=1.0, step=0.01, label='Denoising strength', value=0.7, elem_id="txt2img_denoising_strength") + + with FormRow(elem_id="txt2img_hires_fix_row2", variant="compact"): + hr_scale = gr.Slider(minimum=1.0, maximum=4.0, step=0.05, label="Upscale by", value=2.0, elem_id="txt2img_hr_scale") + hr_resize_x = gr.Slider(minimum=0, maximum=2048, step=8, label="Resize width to", value=0, elem_id="txt2img_hr_resize_x") + hr_resize_y = gr.Slider(minimum=0, maximum=2048, step=8, label="Resize height to", value=0, elem_id="txt2img_hr_resize_y") + + with FormRow(elem_id="txt2img_hires_fix_row3", variant="compact", visible=opts.hires_fix_show_sampler) as hr_sampler_container: + + hr_checkpoint_name = gr.Dropdown(label='Hires checkpoint', elem_id="hr_checkpoint", choices=["Use same checkpoint"] + modules.sd_models.checkpoint_tiles(use_short=True), value="Use same checkpoint") + create_refresh_button(hr_checkpoint_name, modules.sd_models.list_models, lambda: {"choices": ["Use same checkpoint"] + modules.sd_models.checkpoint_tiles(use_short=True)}, "hr_checkpoint_refresh") + + hr_sampler_name = gr.Dropdown(label='Hires sampling method', elem_id="hr_sampler", choices=["Use same sampler"] + sd_samplers.visible_sampler_names(), value="Use same sampler") + + with FormRow(elem_id="txt2img_hires_fix_row4", variant="compact", visible=opts.hires_fix_show_prompts) as hr_prompts_container: + with gr.Column(scale=80): + with gr.Row(): + hr_prompt = gr.Textbox(label="Hires prompt", elem_id="hires_prompt", show_label=False, lines=3, placeholder="Prompt for hires fix pass.\nLeave empty to use the same prompt as in first pass.", elem_classes=["prompt"]) + with gr.Column(scale=80): + with gr.Row(): + hr_negative_prompt = gr.Textbox(label="Hires negative prompt", elem_id="hires_neg_prompt", show_label=False, lines=3, placeholder="Negative prompt for hires fix pass.\nLeave empty to use the same negative prompt as in first pass.", elem_classes=["prompt"]) + + scripts.scripts_txt2img.setup_ui_for_section(category) + + elif category == "batch": + if not opts.dimensions_and_batch_together: + with FormRow(elem_id="txt2img_column_batch"): batch_count = gr.Slider(minimum=1, step=1, label='Batch count', value=1, elem_id="txt2img_batch_count") batch_size = gr.Slider(minimum=1, maximum=8, step=1, label='Batch size', value=1, elem_id="txt2img_batch_size") - elif category == "cfg": - with gr.Row(): - cfg_scale = gr.Slider(minimum=1.0, maximum=30.0, step=0.5, label='CFG Scale', value=7.0, elem_id="txt2img_cfg_scale") + elif category == "override_settings": + with FormRow(elem_id="txt2img_override_settings_row") as row: + override_settings = create_override_settings_dropdown('txt2img', row) - elif category == "checkboxes": - with FormRow(elem_classes="checkboxes-row", variant="compact"): - pass - - elif category == "accordions": - with gr.Row(elem_id="txt2img_accordions", elem_classes="accordions"): - with InputAccordion(False, label="Hires. fix", elem_id="txt2img_hr") as enable_hr: - with enable_hr.extra(): - hr_final_resolution = FormHTML(value="", elem_id="txtimg_hr_finalres", label="Upscaled resolution", interactive=False, min_width=0) - - with FormRow(elem_id="txt2img_hires_fix_row1", variant="compact"): - hr_upscaler = gr.Dropdown(label="Upscaler", elem_id="txt2img_hr_upscaler", choices=[*shared.latent_upscale_modes, *[x.name for x in shared.sd_upscalers]], value=shared.latent_upscale_default_mode) - hr_second_pass_steps = gr.Slider(minimum=0, maximum=150, step=1, label='Hires steps', value=0, elem_id="txt2img_hires_steps") - denoising_strength = gr.Slider(minimum=0.0, maximum=1.0, step=0.01, label='Denoising strength', value=0.7, elem_id="txt2img_denoising_strength") - - with FormRow(elem_id="txt2img_hires_fix_row2", variant="compact"): - hr_scale = gr.Slider(minimum=1.0, maximum=4.0, step=0.05, label="Upscale by", value=2.0, elem_id="txt2img_hr_scale") - hr_resize_x = gr.Slider(minimum=0, maximum=2048, step=8, label="Resize width to", value=0, elem_id="txt2img_hr_resize_x") - hr_resize_y = gr.Slider(minimum=0, maximum=2048, step=8, label="Resize height to", value=0, elem_id="txt2img_hr_resize_y") - - with FormRow(elem_id="txt2img_hires_fix_row3", variant="compact", visible=opts.hires_fix_show_sampler) as hr_sampler_container: - - hr_checkpoint_name = gr.Dropdown(label='Hires checkpoint', elem_id="hr_checkpoint", choices=["Use same checkpoint"] + modules.sd_models.checkpoint_tiles(use_short=True), value="Use same checkpoint") - create_refresh_button(hr_checkpoint_name, modules.sd_models.list_models, lambda: {"choices": ["Use same checkpoint"] + modules.sd_models.checkpoint_tiles(use_short=True)}, "hr_checkpoint_refresh") - - hr_sampler_name = gr.Dropdown(label='Hires sampling method', elem_id="hr_sampler", choices=["Use same sampler"] + sd_samplers.visible_sampler_names(), value="Use same sampler") - - with FormRow(elem_id="txt2img_hires_fix_row4", variant="compact", visible=opts.hires_fix_show_prompts) as hr_prompts_container: - with gr.Column(scale=80): - with gr.Row(): - hr_prompt = gr.Textbox(label="Hires prompt", elem_id="hires_prompt", show_label=False, lines=3, placeholder="Prompt for hires fix pass.\nLeave empty to use the same prompt as in first pass.", elem_classes=["prompt"]) - with gr.Column(scale=80): - with gr.Row(): - hr_negative_prompt = gr.Textbox(label="Hires negative prompt", elem_id="hires_neg_prompt", show_label=False, lines=3, placeholder="Negative prompt for hires fix pass.\nLeave empty to use the same negative prompt as in first pass.", elem_classes=["prompt"]) + elif category == "scripts": + with FormGroup(elem_id="txt2img_script_container"): + custom_inputs = scripts.scripts_txt2img.setup_ui() + if category not in {"accordions"}: scripts.scripts_txt2img.setup_ui_for_section(category) - elif category == "batch": - if not opts.dimensions_and_batch_together: - with FormRow(elem_id="txt2img_column_batch"): - batch_count = gr.Slider(minimum=1, step=1, label='Batch count', value=1, elem_id="txt2img_batch_count") - batch_size = gr.Slider(minimum=1, maximum=8, step=1, label='Batch size', value=1, elem_id="txt2img_batch_size") - - elif category == "override_settings": - with FormRow(elem_id="txt2img_override_settings_row") as row: - override_settings = create_override_settings_dropdown('txt2img', row) - - elif category == "scripts": - with FormGroup(elem_id="txt2img_script_container"): - custom_inputs = scripts.scripts_txt2img.setup_ui() - - if category not in {"accordions"}: - scripts.scripts_txt2img.setup_ui_for_section(category) - hr_resolution_preview_inputs = [enable_hr, width, height, hr_scale, hr_resize_x, hr_resize_y] for component in hr_resolution_preview_inputs: @@ -560,215 +561,216 @@ def create_ui(): extra_tabs.__enter__() with gr.Tab("Generation", id="img2img_generation") as img2img_generation_tab, ResizeHandleRow(equal_height=False): - with gr.Column(variant='compact', elem_id="img2img_settings"): - copy_image_buttons = [] - copy_image_destinations = {} + with gr.Accordion("Open for Settings", open=False) if shared.opts.img2img_settings_accordion else gr.Group(): + with gr.Column(variant='compact', elem_id="img2img_settings"): + copy_image_buttons = [] + copy_image_destinations = {} - def add_copy_image_controls(tab_name, elem): - with gr.Row(variant="compact", elem_id=f"img2img_copy_to_{tab_name}"): - gr.HTML("Copy image to: ", elem_id=f"img2img_label_copy_to_{tab_name}") + def add_copy_image_controls(tab_name, elem): + with gr.Row(variant="compact", elem_id=f"img2img_copy_to_{tab_name}"): + gr.HTML("Copy image to: ", elem_id=f"img2img_label_copy_to_{tab_name}") - for title, name in zip(['img2img', 'sketch', 'inpaint', 'inpaint sketch'], ['img2img', 'sketch', 'inpaint', 'inpaint_sketch']): - if name == tab_name: - gr.Button(title, interactive=False) - copy_image_destinations[name] = elem - continue + for title, name in zip(['img2img', 'sketch', 'inpaint', 'inpaint sketch'], ['img2img', 'sketch', 'inpaint', 'inpaint_sketch']): + if name == tab_name: + gr.Button(title, interactive=False) + copy_image_destinations[name] = elem + continue - button = gr.Button(title) - copy_image_buttons.append((button, name, elem)) + button = gr.Button(title) + copy_image_buttons.append((button, name, elem)) - with gr.Tabs(elem_id="mode_img2img"): - img2img_selected_tab = gr.State(0) + with gr.Tabs(elem_id="mode_img2img"): + img2img_selected_tab = gr.State(0) - with gr.TabItem('img2img', id='img2img', elem_id="img2img_img2img_tab") as tab_img2img: - init_img = gr.Image(label="Image for img2img", elem_id="img2img_image", show_label=False, source="upload", interactive=True, type="pil", tool="editor", image_mode="RGBA", height=opts.img2img_editor_height) - add_copy_image_controls('img2img', init_img) + with gr.TabItem('img2img', id='img2img', elem_id="img2img_img2img_tab") as tab_img2img: + init_img = gr.Image(label="Image for img2img", elem_id="img2img_image", show_label=False, source="upload", interactive=True, type="pil", tool="editor", image_mode="RGBA", height=opts.img2img_editor_height) + add_copy_image_controls('img2img', init_img) - with gr.TabItem('Sketch', id='img2img_sketch', elem_id="img2img_img2img_sketch_tab") as tab_sketch: - sketch = gr.Image(label="Image for img2img", elem_id="img2img_sketch", show_label=False, source="upload", interactive=True, type="pil", tool="color-sketch", image_mode="RGB", height=opts.img2img_editor_height, brush_color=opts.img2img_sketch_default_brush_color) - add_copy_image_controls('sketch', sketch) + with gr.TabItem('Sketch', id='img2img_sketch', elem_id="img2img_img2img_sketch_tab") as tab_sketch: + sketch = gr.Image(label="Image for img2img", elem_id="img2img_sketch", show_label=False, source="upload", interactive=True, type="pil", tool="color-sketch", image_mode="RGB", height=opts.img2img_editor_height, brush_color=opts.img2img_sketch_default_brush_color) + add_copy_image_controls('sketch', sketch) - with gr.TabItem('Inpaint', id='inpaint', elem_id="img2img_inpaint_tab") as tab_inpaint: - init_img_with_mask = gr.Image(label="Image for inpainting with mask", show_label=False, elem_id="img2maskimg", source="upload", interactive=True, type="pil", tool="sketch", image_mode="RGBA", height=opts.img2img_editor_height, brush_color=opts.img2img_inpaint_mask_brush_color) - add_copy_image_controls('inpaint', init_img_with_mask) + with gr.TabItem('Inpaint', id='inpaint', elem_id="img2img_inpaint_tab") as tab_inpaint: + init_img_with_mask = gr.Image(label="Image for inpainting with mask", show_label=False, elem_id="img2maskimg", source="upload", interactive=True, type="pil", tool="sketch", image_mode="RGBA", height=opts.img2img_editor_height, brush_color=opts.img2img_inpaint_mask_brush_color) + add_copy_image_controls('inpaint', init_img_with_mask) - with gr.TabItem('Inpaint sketch', id='inpaint_sketch', elem_id="img2img_inpaint_sketch_tab") as tab_inpaint_color: - inpaint_color_sketch = gr.Image(label="Color sketch inpainting", show_label=False, elem_id="inpaint_sketch", source="upload", interactive=True, type="pil", tool="color-sketch", image_mode="RGB", height=opts.img2img_editor_height, brush_color=opts.img2img_inpaint_sketch_default_brush_color) - inpaint_color_sketch_orig = gr.State(None) - add_copy_image_controls('inpaint_sketch', inpaint_color_sketch) + with gr.TabItem('Inpaint sketch', id='inpaint_sketch', elem_id="img2img_inpaint_sketch_tab") as tab_inpaint_color: + inpaint_color_sketch = gr.Image(label="Color sketch inpainting", show_label=False, elem_id="inpaint_sketch", source="upload", interactive=True, type="pil", tool="color-sketch", image_mode="RGB", height=opts.img2img_editor_height, brush_color=opts.img2img_inpaint_sketch_default_brush_color) + inpaint_color_sketch_orig = gr.State(None) + add_copy_image_controls('inpaint_sketch', inpaint_color_sketch) - def update_orig(image, state): - if image is not None: - same_size = state is not None and state.size == image.size - has_exact_match = np.any(np.all(np.array(image) == np.array(state), axis=-1)) - edited = same_size and has_exact_match - return image if not edited or state is None else state + def update_orig(image, state): + if image is not None: + same_size = state is not None and state.size == image.size + has_exact_match = np.any(np.all(np.array(image) == np.array(state), axis=-1)) + edited = same_size and has_exact_match + return image if not edited or state is None else state - inpaint_color_sketch.change(update_orig, [inpaint_color_sketch, inpaint_color_sketch_orig], inpaint_color_sketch_orig) + inpaint_color_sketch.change(update_orig, [inpaint_color_sketch, inpaint_color_sketch_orig], inpaint_color_sketch_orig) - with gr.TabItem('Inpaint upload', id='inpaint_upload', elem_id="img2img_inpaint_upload_tab") as tab_inpaint_upload: - init_img_inpaint = gr.Image(label="Image for img2img", show_label=False, source="upload", interactive=True, type="pil", elem_id="img_inpaint_base") - init_mask_inpaint = gr.Image(label="Mask", source="upload", interactive=True, type="pil", image_mode="RGBA", elem_id="img_inpaint_mask") + with gr.TabItem('Inpaint upload', id='inpaint_upload', elem_id="img2img_inpaint_upload_tab") as tab_inpaint_upload: + init_img_inpaint = gr.Image(label="Image for img2img", show_label=False, source="upload", interactive=True, type="pil", elem_id="img_inpaint_base") + init_mask_inpaint = gr.Image(label="Mask", source="upload", interactive=True, type="pil", image_mode="RGBA", elem_id="img_inpaint_mask") - with gr.TabItem('Batch', id='batch', elem_id="img2img_batch_tab") as tab_batch: - hidden = '
Disabled when launched with --hide-ui-dir-config.' if shared.cmd_opts.hide_ui_dir_config else '' - gr.HTML( - "

Process images in a directory on the same machine where the server is running." + - "
Use an empty output directory to save pictures normally instead of writing to the output directory." + - f"
Add inpaint batch mask directory to enable inpaint batch processing." - f"{hidden}

" + with gr.TabItem('Batch', id='batch', elem_id="img2img_batch_tab") as tab_batch: + hidden = '
Disabled when launched with --hide-ui-dir-config.' if shared.cmd_opts.hide_ui_dir_config else '' + gr.HTML( + "

Process images in a directory on the same machine where the server is running." + + "
Use an empty output directory to save pictures normally instead of writing to the output directory." + + f"
Add inpaint batch mask directory to enable inpaint batch processing." + f"{hidden}

" + ) + img2img_batch_input_dir = gr.Textbox(label="Input directory", **shared.hide_dirs, elem_id="img2img_batch_input_dir") + img2img_batch_output_dir = gr.Textbox(label="Output directory", **shared.hide_dirs, elem_id="img2img_batch_output_dir") + img2img_batch_inpaint_mask_dir = gr.Textbox(label="Inpaint batch mask directory (required for inpaint batch processing only)", **shared.hide_dirs, elem_id="img2img_batch_inpaint_mask_dir") + with gr.Accordion("PNG info", open=False): + img2img_batch_use_png_info = gr.Checkbox(label="Append png info to prompts", **shared.hide_dirs, elem_id="img2img_batch_use_png_info") + img2img_batch_png_info_dir = gr.Textbox(label="PNG info directory", **shared.hide_dirs, placeholder="Leave empty to use input directory", elem_id="img2img_batch_png_info_dir") + img2img_batch_png_info_props = gr.CheckboxGroup(["Prompt", "Negative prompt", "Seed", "CFG scale", "Sampler", "Steps", "Model hash"], label="Parameters to take from png info", info="Prompts from png info will be appended to prompts set in ui.") + + img2img_tabs = [tab_img2img, tab_sketch, tab_inpaint, tab_inpaint_color, tab_inpaint_upload, tab_batch] + + for i, tab in enumerate(img2img_tabs): + tab.select(fn=lambda tabnum=i: tabnum, inputs=[], outputs=[img2img_selected_tab]) + + def copy_image(img): + if isinstance(img, dict) and 'image' in img: + return img['image'] + + return img + + for button, name, elem in copy_image_buttons: + button.click( + fn=copy_image, + inputs=[elem], + outputs=[copy_image_destinations[name]], + ) + button.click( + fn=lambda: None, + _js=f"switch_to_{name.replace(' ', '_')}", + inputs=[], + outputs=[], ) - img2img_batch_input_dir = gr.Textbox(label="Input directory", **shared.hide_dirs, elem_id="img2img_batch_input_dir") - img2img_batch_output_dir = gr.Textbox(label="Output directory", **shared.hide_dirs, elem_id="img2img_batch_output_dir") - img2img_batch_inpaint_mask_dir = gr.Textbox(label="Inpaint batch mask directory (required for inpaint batch processing only)", **shared.hide_dirs, elem_id="img2img_batch_inpaint_mask_dir") - with gr.Accordion("PNG info", open=False): - img2img_batch_use_png_info = gr.Checkbox(label="Append png info to prompts", **shared.hide_dirs, elem_id="img2img_batch_use_png_info") - img2img_batch_png_info_dir = gr.Textbox(label="PNG info directory", **shared.hide_dirs, placeholder="Leave empty to use input directory", elem_id="img2img_batch_png_info_dir") - img2img_batch_png_info_props = gr.CheckboxGroup(["Prompt", "Negative prompt", "Seed", "CFG scale", "Sampler", "Steps", "Model hash"], label="Parameters to take from png info", info="Prompts from png info will be appended to prompts set in ui.") - img2img_tabs = [tab_img2img, tab_sketch, tab_inpaint, tab_inpaint_color, tab_inpaint_upload, tab_batch] + with FormRow(): + resize_mode = gr.Radio(label="Resize mode", elem_id="resize_mode", choices=["Just resize", "Crop and resize", "Resize and fill", "Just resize (latent upscale)"], type="index", value="Just resize") - for i, tab in enumerate(img2img_tabs): - tab.select(fn=lambda tabnum=i: tabnum, inputs=[], outputs=[img2img_selected_tab]) + scripts.scripts_img2img.prepare_ui() - def copy_image(img): - if isinstance(img, dict) and 'image' in img: - return img['image'] + for category in ordered_ui_categories(): + if category == "sampler": + steps, sampler_name = create_sampler_and_steps_selection(sd_samplers.visible_sampler_names(), "img2img") - return img + elif category == "dimensions": + with FormRow(): + with gr.Column(elem_id="img2img_column_size", scale=4): + selected_scale_tab = gr.State(value=0) - for button, name, elem in copy_image_buttons: - button.click( - fn=copy_image, - inputs=[elem], - outputs=[copy_image_destinations[name]], - ) - button.click( - fn=lambda: None, - _js=f"switch_to_{name.replace(' ', '_')}", - inputs=[], - outputs=[], - ) + with gr.Tabs(): + with gr.Tab(label="Resize to", elem_id="img2img_tab_resize_to") as tab_scale_to: + with FormRow(): + with gr.Column(elem_id="img2img_column_size", scale=4): + width = gr.Slider(minimum=64, maximum=2048, step=8, label="Width", value=512, elem_id="img2img_width") + height = gr.Slider(minimum=64, maximum=2048, step=8, label="Height", value=512, elem_id="img2img_height") + with gr.Column(elem_id="img2img_dimensions_row", scale=1, elem_classes="dimensions-tools"): + res_switch_btn = ToolButton(value=switch_values_symbol, elem_id="img2img_res_switch_btn", tooltip="Switch width/height") + detect_image_size_btn = ToolButton(value=detect_image_size_symbol, elem_id="img2img_detect_image_size_btn", tooltip="Auto detect size from img2img") - with FormRow(): - resize_mode = gr.Radio(label="Resize mode", elem_id="resize_mode", choices=["Just resize", "Crop and resize", "Resize and fill", "Just resize (latent upscale)"], type="index", value="Just resize") + with gr.Tab(label="Resize by", elem_id="img2img_tab_resize_by") as tab_scale_by: + scale_by = gr.Slider(minimum=0.05, maximum=4.0, step=0.05, label="Scale", value=1.0, elem_id="img2img_scale") - scripts.scripts_img2img.prepare_ui() + with FormRow(): + scale_by_html = FormHTML(resize_from_to_html(0, 0, 0.0), elem_id="img2img_scale_resolution_preview") + gr.Slider(label="Unused", elem_id="img2img_unused_scale_by_slider") + button_update_resize_to = gr.Button(visible=False, elem_id="img2img_update_resize_to") - for category in ordered_ui_categories(): - if category == "sampler": - steps, sampler_name = create_sampler_and_steps_selection(sd_samplers.visible_sampler_names(), "img2img") + on_change_args = dict( + fn=resize_from_to_html, + _js="currentImg2imgSourceResolution", + inputs=[dummy_component, dummy_component, scale_by], + outputs=scale_by_html, + show_progress=False, + ) - elif category == "dimensions": - with FormRow(): - with gr.Column(elem_id="img2img_column_size", scale=4): - selected_scale_tab = gr.State(value=0) + scale_by.release(**on_change_args) + button_update_resize_to.click(**on_change_args) - with gr.Tabs(): - with gr.Tab(label="Resize to", elem_id="img2img_tab_resize_to") as tab_scale_to: - with FormRow(): - with gr.Column(elem_id="img2img_column_size", scale=4): - width = gr.Slider(minimum=64, maximum=2048, step=8, label="Width", value=512, elem_id="img2img_width") - height = gr.Slider(minimum=64, maximum=2048, step=8, label="Height", value=512, elem_id="img2img_height") - with gr.Column(elem_id="img2img_dimensions_row", scale=1, elem_classes="dimensions-tools"): - res_switch_btn = ToolButton(value=switch_values_symbol, elem_id="img2img_res_switch_btn", tooltip="Switch width/height") - detect_image_size_btn = ToolButton(value=detect_image_size_symbol, elem_id="img2img_detect_image_size_btn", tooltip="Auto detect size from img2img") + # the code below is meant to update the resolution label after the image in the image selection UI has changed. + # as it is now the event keeps firing continuously for inpaint edits, which ruins the page with constant requests. + # I assume this must be a gradio bug and for now we'll just do it for non-inpaint inputs. + for component in [init_img, sketch]: + component.change(fn=lambda: None, _js="updateImg2imgResizeToTextAfterChangingImage", inputs=[], outputs=[], show_progress=False) - with gr.Tab(label="Resize by", elem_id="img2img_tab_resize_by") as tab_scale_by: - scale_by = gr.Slider(minimum=0.05, maximum=4.0, step=0.05, label="Scale", value=1.0, elem_id="img2img_scale") + tab_scale_to.select(fn=lambda: 0, inputs=[], outputs=[selected_scale_tab]) + tab_scale_by.select(fn=lambda: 1, inputs=[], outputs=[selected_scale_tab]) - with FormRow(): - scale_by_html = FormHTML(resize_from_to_html(0, 0, 0.0), elem_id="img2img_scale_resolution_preview") - gr.Slider(label="Unused", elem_id="img2img_unused_scale_by_slider") - button_update_resize_to = gr.Button(visible=False, elem_id="img2img_update_resize_to") + if opts.dimensions_and_batch_together: + with gr.Column(elem_id="img2img_column_batch"): + batch_count = gr.Slider(minimum=1, step=1, label='Batch count', value=1, elem_id="img2img_batch_count") + batch_size = gr.Slider(minimum=1, maximum=8, step=1, label='Batch size', value=1, elem_id="img2img_batch_size") - on_change_args = dict( - fn=resize_from_to_html, - _js="currentImg2imgSourceResolution", - inputs=[dummy_component, dummy_component, scale_by], - outputs=scale_by_html, - show_progress=False, - ) + elif category == "denoising": + denoising_strength = gr.Slider(minimum=0.0, maximum=1.0, step=0.01, label='Denoising strength', value=0.75, elem_id="img2img_denoising_strength") - scale_by.release(**on_change_args) - button_update_resize_to.click(**on_change_args) + elif category == "cfg": + with gr.Row(): + cfg_scale = gr.Slider(minimum=1.0, maximum=30.0, step=0.5, label='CFG Scale', value=7.0, elem_id="img2img_cfg_scale") + image_cfg_scale = gr.Slider(minimum=0, maximum=3.0, step=0.05, label='Image CFG Scale', value=1.5, elem_id="img2img_image_cfg_scale", visible=False) - # the code below is meant to update the resolution label after the image in the image selection UI has changed. - # as it is now the event keeps firing continuously for inpaint edits, which ruins the page with constant requests. - # I assume this must be a gradio bug and for now we'll just do it for non-inpaint inputs. - for component in [init_img, sketch]: - component.change(fn=lambda: None, _js="updateImg2imgResizeToTextAfterChangingImage", inputs=[], outputs=[], show_progress=False) + elif category == "checkboxes": + with FormRow(elem_classes="checkboxes-row", variant="compact"): + pass - tab_scale_to.select(fn=lambda: 0, inputs=[], outputs=[selected_scale_tab]) - tab_scale_by.select(fn=lambda: 1, inputs=[], outputs=[selected_scale_tab]) + elif category == "accordions": + with gr.Row(elem_id="img2img_accordions", elem_classes="accordions"): + scripts.scripts_img2img.setup_ui_for_section(category) - if opts.dimensions_and_batch_together: - with gr.Column(elem_id="img2img_column_batch"): + elif category == "batch": + if not opts.dimensions_and_batch_together: + with FormRow(elem_id="img2img_column_batch"): batch_count = gr.Slider(minimum=1, step=1, label='Batch count', value=1, elem_id="img2img_batch_count") batch_size = gr.Slider(minimum=1, maximum=8, step=1, label='Batch size', value=1, elem_id="img2img_batch_size") - elif category == "denoising": - denoising_strength = gr.Slider(minimum=0.0, maximum=1.0, step=0.01, label='Denoising strength', value=0.75, elem_id="img2img_denoising_strength") + elif category == "override_settings": + with FormRow(elem_id="img2img_override_settings_row") as row: + override_settings = create_override_settings_dropdown('img2img', row) - elif category == "cfg": - with gr.Row(): - cfg_scale = gr.Slider(minimum=1.0, maximum=30.0, step=0.5, label='CFG Scale', value=7.0, elem_id="img2img_cfg_scale") - image_cfg_scale = gr.Slider(minimum=0, maximum=3.0, step=0.05, label='Image CFG Scale', value=1.5, elem_id="img2img_image_cfg_scale", visible=False) + elif category == "scripts": + with FormGroup(elem_id="img2img_script_container"): + custom_inputs = scripts.scripts_img2img.setup_ui() - elif category == "checkboxes": - with FormRow(elem_classes="checkboxes-row", variant="compact"): - pass + elif category == "inpaint": + with FormGroup(elem_id="inpaint_controls", visible=False) as inpaint_controls: + with FormRow(): + mask_blur = gr.Slider(label='Mask blur', minimum=0, maximum=64, step=1, value=4, elem_id="img2img_mask_blur") + mask_alpha = gr.Slider(label="Mask transparency", visible=False, elem_id="img2img_mask_alpha") - elif category == "accordions": - with gr.Row(elem_id="img2img_accordions", elem_classes="accordions"): + with FormRow(): + inpainting_mask_invert = gr.Radio(label='Mask mode', choices=['Inpaint masked', 'Inpaint not masked'], value='Inpaint masked', type="index", elem_id="img2img_mask_mode") + + with FormRow(): + inpainting_fill = gr.Radio(label='Masked content', choices=['fill', 'original', 'latent noise', 'latent nothing'], value='original', type="index", elem_id="img2img_inpainting_fill") + + with FormRow(): + with gr.Column(): + inpaint_full_res = gr.Radio(label="Inpaint area", choices=["Whole picture", "Only masked"], type="index", value="Whole picture", elem_id="img2img_inpaint_full_res") + + with gr.Column(scale=4): + inpaint_full_res_padding = gr.Slider(label='Only masked padding, pixels', minimum=0, maximum=256, step=4, value=32, elem_id="img2img_inpaint_full_res_padding") + + def select_img2img_tab(tab): + return gr.update(visible=tab in [2, 3, 4]), gr.update(visible=tab == 3), + + for i, elem in enumerate(img2img_tabs): + elem.select( + fn=lambda tab=i: select_img2img_tab(tab), + inputs=[], + outputs=[inpaint_controls, mask_alpha], + ) + + if category not in {"accordions"}: scripts.scripts_img2img.setup_ui_for_section(category) - elif category == "batch": - if not opts.dimensions_and_batch_together: - with FormRow(elem_id="img2img_column_batch"): - batch_count = gr.Slider(minimum=1, step=1, label='Batch count', value=1, elem_id="img2img_batch_count") - batch_size = gr.Slider(minimum=1, maximum=8, step=1, label='Batch size', value=1, elem_id="img2img_batch_size") - - elif category == "override_settings": - with FormRow(elem_id="img2img_override_settings_row") as row: - override_settings = create_override_settings_dropdown('img2img', row) - - elif category == "scripts": - with FormGroup(elem_id="img2img_script_container"): - custom_inputs = scripts.scripts_img2img.setup_ui() - - elif category == "inpaint": - with FormGroup(elem_id="inpaint_controls", visible=False) as inpaint_controls: - with FormRow(): - mask_blur = gr.Slider(label='Mask blur', minimum=0, maximum=64, step=1, value=4, elem_id="img2img_mask_blur") - mask_alpha = gr.Slider(label="Mask transparency", visible=False, elem_id="img2img_mask_alpha") - - with FormRow(): - inpainting_mask_invert = gr.Radio(label='Mask mode', choices=['Inpaint masked', 'Inpaint not masked'], value='Inpaint masked', type="index", elem_id="img2img_mask_mode") - - with FormRow(): - inpainting_fill = gr.Radio(label='Masked content', choices=['fill', 'original', 'latent noise', 'latent nothing'], value='original', type="index", elem_id="img2img_inpainting_fill") - - with FormRow(): - with gr.Column(): - inpaint_full_res = gr.Radio(label="Inpaint area", choices=["Whole picture", "Only masked"], type="index", value="Whole picture", elem_id="img2img_inpaint_full_res") - - with gr.Column(scale=4): - inpaint_full_res_padding = gr.Slider(label='Only masked padding, pixels', minimum=0, maximum=256, step=4, value=32, elem_id="img2img_inpaint_full_res_padding") - - def select_img2img_tab(tab): - return gr.update(visible=tab in [2, 3, 4]), gr.update(visible=tab == 3), - - for i, elem in enumerate(img2img_tabs): - elem.select( - fn=lambda tab=i: select_img2img_tab(tab), - inputs=[], - outputs=[inpaint_controls, mask_alpha], - ) - - if category not in {"accordions"}: - scripts.scripts_img2img.setup_ui_for_section(category) - img2img_gallery, generation_info, html_info, html_log = create_output_panel("img2img", opts.outdir_img2img_samples) img2img_args = dict( From fe1967a4c4a02eccfa45b65ee19a5b0773ced31c Mon Sep 17 00:00:00 2001 From: v0xie <28695009+v0xie@users.noreply.github.com> Date: Fri, 3 Nov 2023 17:52:55 -0700 Subject: [PATCH 025/139] skip multihead attn for now --- extensions-builtin/Lora/network_oft.py | 52 ++++++++++++++++++-------- 1 file changed, 36 insertions(+), 16 deletions(-) diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py index e102eafc1..979a20476 100644 --- a/extensions-builtin/Lora/network_oft.py +++ b/extensions-builtin/Lora/network_oft.py @@ -18,6 +18,7 @@ class NetworkModuleOFT(network.NetworkModule): super().__init__(net, weights) self.lin_module = None + self.org_module: list[torch.Module] = [self.sd_module] # kohya-ss if "oft_blocks" in weights.w.keys(): self.is_kohya = True @@ -30,7 +31,7 @@ class NetworkModuleOFT(network.NetworkModule): # alpha is rank if alpha is 0 or None if self.alpha is None: pass - self.dim = self.oft_blocks.shape[0] # FIXME: almost certainly incorrect, assumes tensor is shape [*, m, n] + self.dim = self.oft_blocks.shape[1] # FIXME: almost certainly incorrect, assumes tensor is shape [*, m, n] else: raise ValueError("oft_blocks or oft_diag must be in weights dict") @@ -46,6 +47,12 @@ class NetworkModuleOFT(network.NetworkModule): # raise ValueError("Linear sd_module must have out_features or embed_dim") elif is_other_linear: self.out_dim = self.sd_module.embed_dim + #self.org_weight = self.org_module[0].weight +# if hasattr(self.sd_module, "in_proj_weight"): +# self.in_proj_dim = self.sd_module.in_proj_weight.shape[1] +# if hasattr(self.sd_module, "out_proj_weight"): +# self.out_proj_dim = self.sd_module.out_proj_weight.shape[0] +# self.in_proj_dim = self.sd_module.in_proj_weight.shape[1] elif is_conv: self.out_dim = self.sd_module.out_channels else: @@ -58,10 +65,9 @@ class NetworkModuleOFT(network.NetworkModule): self.constraint = self.alpha * self.out_dim #elif is_linear or is_conv: else: - self.num_blocks, self.block_size = factorization(self.out_dim, self.dim) + self.block_size, self.num_blocks = factorization(self.out_dim, self.dim) self.constraint = None - self.org_module: list[torch.Module] = [self.sd_module] # if is_other_linear: # weight = self.oft_blocks.reshape(self.oft_blocks.shape[0], -1) @@ -110,25 +116,39 @@ class NetworkModuleOFT(network.NetworkModule): def calc_updown(self, orig_weight): multiplier = self.multiplier() * self.calc_scale() - R = self.get_weight(self.oft_blocks, multiplier) - #R = self.oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) - merged_weight = self.merge_weight(R, orig_weight) + is_other_linear = type(self.sd_module) in [ torch.nn.MultiheadAttention] + if self.is_kohya and not is_other_linear: + R = self.get_weight(self.oft_blocks, multiplier) + #R = self.oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) + merged_weight = self.merge_weight(R, orig_weight) + elif not self.is_kohya and not is_other_linear: + if is_other_linear and orig_weight.shape[0] != orig_weight.shape[1]: + orig_weight=orig_weight.permute(1, 0) + R = self.oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) + merged_weight = rearrange(orig_weight, '(k n) ... -> k n ...', k=self.num_blocks, n=self.block_size) + #orig_weight = rearrange(orig_weight, '(k n) ... -> k n ...', k=self.block_size, n=self.num_blocks) + merged_weight = torch.einsum( + 'k n m, k n ... -> k m ...', + R * multiplier + torch.eye(self.block_size, device=orig_weight.device), + merged_weight + ) + merged_weight = rearrange(merged_weight, 'k m ... -> (k m) ...') + if is_other_linear and orig_weight.shape[0] != orig_weight.shape[1]: + orig_weight=orig_weight.permute(1, 0) + #merged_weight=merged_weight.permute(1, 0) + updown = merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight + #updown = weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight + output_shape = orig_weight.shape + else: + # skip for now + updown = torch.zeros([orig_weight.shape[1], orig_weight.shape[1]], device=orig_weight.device, dtype=orig_weight.dtype) + output_shape = (orig_weight.shape[1], orig_weight.shape[1]) #if self.lin_module is not None: # R = self.lin_module.weight.to(orig_weight.device, dtype=orig_weight.dtype) # weight = torch.mul(torch.mul(R, multiplier), orig_weight) #else: - # orig_weight = rearrange(orig_weight, '(k n) ... -> k n ...', k=self.num_blocks, n=self.block_size) - # weight = torch.einsum( - # 'k n m, k n ... -> k m ...', - # R * multiplier + torch.eye(self.block_size, device=orig_weight.device), - # orig_weight - # ) - # weight = rearrange(weight, 'k m ... -> (k m) ...') - updown = merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight - #updown = weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight - output_shape = orig_weight.shape orig_weight = orig_weight return self.finalize_updown(updown, orig_weight, output_shape) From f6c8201e5663ca2182a66c8eca63ce4801d52849 Mon Sep 17 00:00:00 2001 From: v0xie <28695009+v0xie@users.noreply.github.com> Date: Fri, 3 Nov 2023 19:35:15 -0700 Subject: [PATCH 026/139] refactor: move factorization to lyco_helpers, separate calc_updown for kohya and kb --- extensions-builtin/Lora/lyco_helpers.py | 47 +++++++++ extensions-builtin/Lora/network_oft.py | 131 ++++++------------------ 2 files changed, 77 insertions(+), 101 deletions(-) diff --git a/extensions-builtin/Lora/lyco_helpers.py b/extensions-builtin/Lora/lyco_helpers.py index 279b34bc9..1679a0ce6 100644 --- a/extensions-builtin/Lora/lyco_helpers.py +++ b/extensions-builtin/Lora/lyco_helpers.py @@ -19,3 +19,50 @@ def rebuild_cp_decomposition(up, down, mid): up = up.reshape(up.size(0), -1) down = down.reshape(down.size(0), -1) return torch.einsum('n m k l, i n, m j -> i j k l', mid, up, down) + + +# copied from https://github.com/KohakuBlueleaf/LyCORIS/blob/dev/lycoris/modules/lokr.py +def factorization(dimension: int, factor:int=-1) -> tuple[int, int]: + ''' + return a tuple of two value of input dimension decomposed by the number closest to factor + second value is higher or equal than first value. + + In LoRA with Kroneckor Product, first value is a value for weight scale. + secon value is a value for weight. + + Becuase of non-commutative property, A⊗B ≠ B⊗A. Meaning of two matrices is slightly different. + + examples) + factor + -1 2 4 8 16 ... + 127 -> 1, 127 127 -> 1, 127 127 -> 1, 127 127 -> 1, 127 127 -> 1, 127 + 128 -> 8, 16 128 -> 2, 64 128 -> 4, 32 128 -> 8, 16 128 -> 8, 16 + 250 -> 10, 25 250 -> 2, 125 250 -> 2, 125 250 -> 5, 50 250 -> 10, 25 + 360 -> 8, 45 360 -> 2, 180 360 -> 4, 90 360 -> 8, 45 360 -> 12, 30 + 512 -> 16, 32 512 -> 2, 256 512 -> 4, 128 512 -> 8, 64 512 -> 16, 32 + 1024 -> 32, 32 1024 -> 2, 512 1024 -> 4, 256 1024 -> 8, 128 1024 -> 16, 64 + ''' + + if factor > 0 and (dimension % factor) == 0: + m = factor + n = dimension // factor + if m > n: + n, m = m, n + return m, n + if factor < 0: + factor = dimension + m, n = 1, dimension + length = m + n + while m length or new_m>factor: + break + else: + m, n = new_m, new_n + if m > n: + n, m = m, n + return m, n + diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py index 979a20476..2be67fe53 100644 --- a/extensions-builtin/Lora/network_oft.py +++ b/extensions-builtin/Lora/network_oft.py @@ -1,7 +1,7 @@ import torch import network +from lyco_helpers import factorization from einops import rearrange -from modules import devices class ModuleTypeOFT(network.ModuleType): @@ -11,7 +11,8 @@ class ModuleTypeOFT(network.ModuleType): return None -# adapted from kohya's implementation https://github.com/kohya-ss/sd-scripts/blob/main/networks/oft.py +# adapted from kohya-ss' implementation https://github.com/kohya-ss/sd-scripts/blob/main/networks/oft.py +# and KohakuBlueleaf's implementation https://github.com/KohakuBlueleaf/LyCORIS/blob/dev/lycoris/modules/diag_oft.py class NetworkModuleOFT(network.NetworkModule): def __init__(self, net: network.Network, weights: network.NetworkWeights): @@ -19,6 +20,7 @@ class NetworkModuleOFT(network.NetworkModule): self.lin_module = None self.org_module: list[torch.Module] = [self.sd_module] + # kohya-ss if "oft_blocks" in weights.w.keys(): self.is_kohya = True @@ -37,61 +39,31 @@ class NetworkModuleOFT(network.NetworkModule): is_linear = type(self.sd_module) in [torch.nn.Linear, torch.nn.modules.linear.NonDynamicallyQuantizableLinear] is_conv = type(self.sd_module) in [torch.nn.Conv2d] - is_other_linear = type(self.sd_module) in [ torch.nn.MultiheadAttention] - #if "Linear" in self.sd_module.__class__.__name__ or is_linear: + is_other_linear = type(self.sd_module) in [torch.nn.MultiheadAttention] + if is_linear: self.out_dim = self.sd_module.out_features - #elif hasattr(self.sd_module, "embed_dim"): - # self.out_dim = self.sd_module.embed_dim - #else: - # raise ValueError("Linear sd_module must have out_features or embed_dim") elif is_other_linear: self.out_dim = self.sd_module.embed_dim - #self.org_weight = self.org_module[0].weight -# if hasattr(self.sd_module, "in_proj_weight"): -# self.in_proj_dim = self.sd_module.in_proj_weight.shape[1] -# if hasattr(self.sd_module, "out_proj_weight"): -# self.out_proj_dim = self.sd_module.out_proj_weight.shape[0] -# self.in_proj_dim = self.sd_module.in_proj_weight.shape[1] elif is_conv: self.out_dim = self.sd_module.out_channels else: raise ValueError("sd_module must be Linear or Conv") - if self.is_kohya: self.num_blocks = self.dim self.block_size = self.out_dim // self.num_blocks self.constraint = self.alpha * self.out_dim - #elif is_linear or is_conv: else: self.block_size, self.num_blocks = factorization(self.out_dim, self.dim) self.constraint = None - - # if is_other_linear: - # weight = self.oft_blocks.reshape(self.oft_blocks.shape[0], -1) - # module = torch.nn.Linear(weight.shape[1], weight.shape[0], bias=False) - # with torch.no_grad(): - # if weight.shape != module.weight.shape: - # weight = weight.reshape(module.weight.shape) - # module.weight.copy_(weight) - # module.to(device=devices.cpu, dtype=devices.dtype) - # module.weight.requires_grad_(False) - # self.lin_module = module - #return module - def merge_weight(self, R_weight, org_weight): R_weight = R_weight.to(org_weight.device, dtype=org_weight.dtype) if org_weight.dim() == 4: weight = torch.einsum("oihw, op -> pihw", org_weight, R_weight) else: weight = torch.einsum("oi, op -> pi", org_weight, R_weight) - #weight = torch.einsum( - # "k n m, k n ... -> k m ...", - # self.oft_diag * scale + torch.eye(self.block_size, device=device), - # org_weight - #) return weight def get_weight(self, oft_blocks, multiplier=None): @@ -111,48 +83,51 @@ class NetworkModuleOFT(network.NetworkModule): block_R_weighted = multiplier * block_R + (1 - multiplier) * m_I R = torch.block_diag(*block_R_weighted) return R - #return self.oft_blocks + def calc_updown_kohya(self, orig_weight, multiplier): + R = self.get_weight(self.oft_blocks, multiplier) + merged_weight = self.merge_weight(R, orig_weight) - def calc_updown(self, orig_weight): - multiplier = self.multiplier() * self.calc_scale() - is_other_linear = type(self.sd_module) in [ torch.nn.MultiheadAttention] - if self.is_kohya and not is_other_linear: - R = self.get_weight(self.oft_blocks, multiplier) - #R = self.oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) - merged_weight = self.merge_weight(R, orig_weight) - elif not self.is_kohya and not is_other_linear: + updown = merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight + output_shape = orig_weight.shape + orig_weight = orig_weight + return self.finalize_updown(updown, orig_weight, output_shape) + + def calc_updown_kb(self, orig_weight, multiplier): + is_other_linear = type(self.sd_module) in [torch.nn.MultiheadAttention] + + if not is_other_linear: if is_other_linear and orig_weight.shape[0] != orig_weight.shape[1]: orig_weight=orig_weight.permute(1, 0) + R = self.oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) merged_weight = rearrange(orig_weight, '(k n) ... -> k n ...', k=self.num_blocks, n=self.block_size) - #orig_weight = rearrange(orig_weight, '(k n) ... -> k n ...', k=self.block_size, n=self.num_blocks) merged_weight = torch.einsum( 'k n m, k n ... -> k m ...', R * multiplier + torch.eye(self.block_size, device=orig_weight.device), - merged_weight + merged_weight ) merged_weight = rearrange(merged_weight, 'k m ... -> (k m) ...') + if is_other_linear and orig_weight.shape[0] != orig_weight.shape[1]: orig_weight=orig_weight.permute(1, 0) - #merged_weight=merged_weight.permute(1, 0) + updown = merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight - #updown = weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight output_shape = orig_weight.shape else: - # skip for now + # FIXME: skip MultiheadAttention for now updown = torch.zeros([orig_weight.shape[1], orig_weight.shape[1]], device=orig_weight.device, dtype=orig_weight.dtype) output_shape = (orig_weight.shape[1], orig_weight.shape[1]) - #if self.lin_module is not None: - # R = self.lin_module.weight.to(orig_weight.device, dtype=orig_weight.dtype) - # weight = torch.mul(torch.mul(R, multiplier), orig_weight) - #else: - - orig_weight = orig_weight - return self.finalize_updown(updown, orig_weight, output_shape) + def calc_updown(self, orig_weight): + multiplier = self.multiplier() * self.calc_scale() + if self.is_kohya: + return self.calc_updown_kohya(orig_weight, multiplier) + else: + return self.calc_updown_kb(orig_weight, multiplier) + # override to remove the multiplier/scale factor; it's already multiplied in get_weight def finalize_updown(self, updown, orig_weight, output_shape, ex_bias=None): #return super().finalize_updown(updown, orig_weight, output_shape, ex_bias) @@ -172,49 +147,3 @@ class NetworkModuleOFT(network.NetworkModule): ex_bias = ex_bias * self.multiplier() return updown, ex_bias - -# copied from https://github.com/KohakuBlueleaf/LyCORIS/blob/dev/lycoris/modules/lokr.py -def factorization(dimension: int, factor:int=-1) -> tuple[int, int]: - ''' - return a tuple of two value of input dimension decomposed by the number closest to factor - second value is higher or equal than first value. - - In LoRA with Kroneckor Product, first value is a value for weight scale. - secon value is a value for weight. - - Becuase of non-commutative property, A⊗B ≠ B⊗A. Meaning of two matrices is slightly different. - - examples) - factor - -1 2 4 8 16 ... - 127 -> 1, 127 127 -> 1, 127 127 -> 1, 127 127 -> 1, 127 127 -> 1, 127 - 128 -> 8, 16 128 -> 2, 64 128 -> 4, 32 128 -> 8, 16 128 -> 8, 16 - 250 -> 10, 25 250 -> 2, 125 250 -> 2, 125 250 -> 5, 50 250 -> 10, 25 - 360 -> 8, 45 360 -> 2, 180 360 -> 4, 90 360 -> 8, 45 360 -> 12, 30 - 512 -> 16, 32 512 -> 2, 256 512 -> 4, 128 512 -> 8, 64 512 -> 16, 32 - 1024 -> 32, 32 1024 -> 2, 512 1024 -> 4, 256 1024 -> 8, 128 1024 -> 16, 64 - ''' - - if factor > 0 and (dimension % factor) == 0: - m = factor - n = dimension // factor - if m > n: - n, m = m, n - return m, n - if factor < 0: - factor = dimension - m, n = 1, dimension - length = m + n - while m length or new_m>factor: - break - else: - m, n = new_m, new_n - if m > n: - n, m = m, n - return m, n - From 329c8bacce706811776e1c1c6a0d39b46886a268 Mon Sep 17 00:00:00 2001 From: v0xie <28695009+v0xie@users.noreply.github.com> Date: Sat, 4 Nov 2023 14:54:36 -0700 Subject: [PATCH 027/139] refactor: use same updown for both kohya OFT and LyCORIS diag-oft --- extensions-builtin/Lora/network_oft.py | 91 +++++++++++++++++++++----- 1 file changed, 74 insertions(+), 17 deletions(-) diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py index 2be67fe53..e4aa082b7 100644 --- a/extensions-builtin/Lora/network_oft.py +++ b/extensions-builtin/Lora/network_oft.py @@ -2,6 +2,7 @@ import torch import network from lyco_helpers import factorization from einops import rearrange +from modules import devices class ModuleTypeOFT(network.ModuleType): @@ -24,12 +25,14 @@ class NetworkModuleOFT(network.NetworkModule): # kohya-ss if "oft_blocks" in weights.w.keys(): self.is_kohya = True - self.oft_blocks = weights.w["oft_blocks"] + self.oft_blocks = weights.w["oft_blocks"] # (num_blocks, block_size, block_size) self.alpha = weights.w["alpha"] - self.dim = self.oft_blocks.shape[0] + self.dim = self.oft_blocks.shape[0] # lora dim + #self.oft_blocks = rearrange(self.oft_blocks, 'k m ... -> (k m) ...') elif "oft_diag" in weights.w.keys(): self.is_kohya = False - self.oft_blocks = weights.w["oft_diag"] + self.oft_blocks = weights.w["oft_diag"] # (num_blocks, block_size, block_size) + # alpha is rank if alpha is 0 or None if self.alpha is None: pass @@ -51,12 +54,57 @@ class NetworkModuleOFT(network.NetworkModule): raise ValueError("sd_module must be Linear or Conv") if self.is_kohya: - self.num_blocks = self.dim - self.block_size = self.out_dim // self.num_blocks + #self.num_blocks = self.dim + #self.block_size = self.out_dim // self.num_blocks + #self.block_size = self.dim + #self.num_blocks = self.out_dim // self.block_size self.constraint = self.alpha * self.out_dim + self.num_blocks, self.block_size = factorization(self.out_dim, self.dim) else: - self.block_size, self.num_blocks = factorization(self.out_dim, self.dim) self.constraint = None + self.block_size, self.num_blocks = factorization(self.out_dim, self.dim) + + if is_other_linear: + self.lin_module = self.create_module(weights.w, "oft_diag", none_ok=True) + + + def create_module(self, weights, key, none_ok=False): + weight = weights.get(key) + + if weight is None and none_ok: + return None + + is_linear = type(self.sd_module) in [torch.nn.Linear, torch.nn.modules.linear.NonDynamicallyQuantizableLinear, torch.nn.MultiheadAttention] + is_conv = type(self.sd_module) in [torch.nn.Conv2d] + + if is_linear: + weight = weight.reshape(weight.shape[0], -1) + module = torch.nn.Linear(weight.shape[1], weight.shape[0], bias=False) + elif is_conv and key == "lora_down.weight" or key == "dyn_up": + if len(weight.shape) == 2: + weight = weight.reshape(weight.shape[0], -1, 1, 1) + + if weight.shape[2] != 1 or weight.shape[3] != 1: + module = torch.nn.Conv2d(weight.shape[1], weight.shape[0], self.sd_module.kernel_size, self.sd_module.stride, self.sd_module.padding, bias=False) + else: + module = torch.nn.Conv2d(weight.shape[1], weight.shape[0], (1, 1), bias=False) + elif is_conv and key == "lora_mid.weight": + module = torch.nn.Conv2d(weight.shape[1], weight.shape[0], self.sd_module.kernel_size, self.sd_module.stride, self.sd_module.padding, bias=False) + elif is_conv and key == "lora_up.weight" or key == "dyn_down": + module = torch.nn.Conv2d(weight.shape[1], weight.shape[0], (1, 1), bias=False) + else: + raise AssertionError(f'Lora layer {self.network_key} matched a layer with unsupported type: {type(self.sd_module).__name__}') + + with torch.no_grad(): + if weight.shape != module.weight.shape: + weight = weight.reshape(module.weight.shape) + module.weight.copy_(weight) + + module.to(device=devices.cpu, dtype=devices.dtype) + module.weight.requires_grad_(False) + + return module + def merge_weight(self, R_weight, org_weight): R_weight = R_weight.to(org_weight.device, dtype=org_weight.dtype) @@ -77,7 +125,8 @@ class NetworkModuleOFT(network.NetworkModule): else: new_norm_Q = norm_Q block_Q = block_Q * ((new_norm_Q + 1e-8) / (norm_Q + 1e-8)) - m_I = torch.eye(self.block_size, device=oft_blocks.device).unsqueeze(0).repeat(self.num_blocks, 1, 1) + m_I = torch.eye(self.num_blocks, device=oft_blocks.device).unsqueeze(0).repeat(self.block_size, 1, 1) + #m_I = torch.eye(self.block_size, device=oft_blocks.device).unsqueeze(0).repeat(self.num_blocks, 1, 1) block_R = torch.matmul(m_I + block_Q, (m_I - block_Q).inverse()) block_R_weighted = multiplier * block_R + (1 - multiplier) * m_I @@ -97,25 +146,33 @@ class NetworkModuleOFT(network.NetworkModule): is_other_linear = type(self.sd_module) in [torch.nn.MultiheadAttention] if not is_other_linear: - if is_other_linear and orig_weight.shape[0] != orig_weight.shape[1]: - orig_weight=orig_weight.permute(1, 0) + #if is_other_linear and orig_weight.shape[0] != orig_weight.shape[1]: + # orig_weight=orig_weight.permute(1, 0) + + oft_blocks = self.oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) + + # without this line the results are significantly worse / less accurate + oft_blocks = oft_blocks - oft_blocks.transpose(1, 2) + + R = oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) + R = R * multiplier + torch.eye(self.block_size, device=orig_weight.device) - R = self.oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) merged_weight = rearrange(orig_weight, '(k n) ... -> k n ...', k=self.num_blocks, n=self.block_size) merged_weight = torch.einsum( 'k n m, k n ... -> k m ...', - R * multiplier + torch.eye(self.block_size, device=orig_weight.device), + R, merged_weight ) merged_weight = rearrange(merged_weight, 'k m ... -> (k m) ...') - if is_other_linear and orig_weight.shape[0] != orig_weight.shape[1]: - orig_weight=orig_weight.permute(1, 0) + #if is_other_linear and orig_weight.shape[0] != orig_weight.shape[1]: + # orig_weight=orig_weight.permute(1, 0) updown = merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight output_shape = orig_weight.shape else: # FIXME: skip MultiheadAttention for now + #up = self.lin_module.weight.to(orig_weight.device, dtype=orig_weight.dtype) updown = torch.zeros([orig_weight.shape[1], orig_weight.shape[1]], device=orig_weight.device, dtype=orig_weight.dtype) output_shape = (orig_weight.shape[1], orig_weight.shape[1]) @@ -123,10 +180,10 @@ class NetworkModuleOFT(network.NetworkModule): def calc_updown(self, orig_weight): multiplier = self.multiplier() * self.calc_scale() - if self.is_kohya: - return self.calc_updown_kohya(orig_weight, multiplier) - else: - return self.calc_updown_kb(orig_weight, multiplier) + #if self.is_kohya: + # return self.calc_updown_kohya(orig_weight, multiplier) + #else: + return self.calc_updown_kb(orig_weight, multiplier) # override to remove the multiplier/scale factor; it's already multiplied in get_weight def finalize_updown(self, updown, orig_weight, output_shape, ex_bias=None): From bbf00a96afb2215f13cc72a7908225ae300c423d Mon Sep 17 00:00:00 2001 From: v0xie <28695009+v0xie@users.noreply.github.com> Date: Sat, 4 Nov 2023 14:56:47 -0700 Subject: [PATCH 028/139] refactor: remove unused function --- extensions-builtin/Lora/network_oft.py | 47 -------------------------- 1 file changed, 47 deletions(-) diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py index e4aa082b7..93402bb28 100644 --- a/extensions-builtin/Lora/network_oft.py +++ b/extensions-builtin/Lora/network_oft.py @@ -2,7 +2,6 @@ import torch import network from lyco_helpers import factorization from einops import rearrange -from modules import devices class ModuleTypeOFT(network.ModuleType): @@ -54,58 +53,12 @@ class NetworkModuleOFT(network.NetworkModule): raise ValueError("sd_module must be Linear or Conv") if self.is_kohya: - #self.num_blocks = self.dim - #self.block_size = self.out_dim // self.num_blocks - #self.block_size = self.dim - #self.num_blocks = self.out_dim // self.block_size self.constraint = self.alpha * self.out_dim self.num_blocks, self.block_size = factorization(self.out_dim, self.dim) else: self.constraint = None self.block_size, self.num_blocks = factorization(self.out_dim, self.dim) - if is_other_linear: - self.lin_module = self.create_module(weights.w, "oft_diag", none_ok=True) - - - def create_module(self, weights, key, none_ok=False): - weight = weights.get(key) - - if weight is None and none_ok: - return None - - is_linear = type(self.sd_module) in [torch.nn.Linear, torch.nn.modules.linear.NonDynamicallyQuantizableLinear, torch.nn.MultiheadAttention] - is_conv = type(self.sd_module) in [torch.nn.Conv2d] - - if is_linear: - weight = weight.reshape(weight.shape[0], -1) - module = torch.nn.Linear(weight.shape[1], weight.shape[0], bias=False) - elif is_conv and key == "lora_down.weight" or key == "dyn_up": - if len(weight.shape) == 2: - weight = weight.reshape(weight.shape[0], -1, 1, 1) - - if weight.shape[2] != 1 or weight.shape[3] != 1: - module = torch.nn.Conv2d(weight.shape[1], weight.shape[0], self.sd_module.kernel_size, self.sd_module.stride, self.sd_module.padding, bias=False) - else: - module = torch.nn.Conv2d(weight.shape[1], weight.shape[0], (1, 1), bias=False) - elif is_conv and key == "lora_mid.weight": - module = torch.nn.Conv2d(weight.shape[1], weight.shape[0], self.sd_module.kernel_size, self.sd_module.stride, self.sd_module.padding, bias=False) - elif is_conv and key == "lora_up.weight" or key == "dyn_down": - module = torch.nn.Conv2d(weight.shape[1], weight.shape[0], (1, 1), bias=False) - else: - raise AssertionError(f'Lora layer {self.network_key} matched a layer with unsupported type: {type(self.sd_module).__name__}') - - with torch.no_grad(): - if weight.shape != module.weight.shape: - weight = weight.reshape(module.weight.shape) - module.weight.copy_(weight) - - module.to(device=devices.cpu, dtype=devices.dtype) - module.weight.requires_grad_(False) - - return module - - def merge_weight(self, R_weight, org_weight): R_weight = R_weight.to(org_weight.device, dtype=org_weight.dtype) if org_weight.dim() == 4: From a625a7bb817cbf6a97d2030dc3a8015a046bd388 Mon Sep 17 00:00:00 2001 From: Emily Zeng Date: Thu, 9 Nov 2023 13:15:06 -0500 Subject: [PATCH 029/139] moved nested with to single line to remove extra tabs --- modules/ui.py | 573 +++++++++++++++++++++++++------------------------- 1 file changed, 286 insertions(+), 287 deletions(-) diff --git a/modules/ui.py b/modules/ui.py index 4a3e60d13..0faccbd34 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -270,89 +270,88 @@ def create_ui(): extra_tabs.__enter__() with gr.Tab("Generation", id="txt2img_generation") as txt2img_generation_tab, ResizeHandleRow(equal_height=False): - with gr.Accordion("Open for Settings", open=False) if shared.opts.img2img_settings_accordion else gr.Group(): - with gr.Column(variant='compact', elem_id="txt2img_settings"): - scripts.scripts_txt2img.prepare_ui() + with gr.Accordion("Open for Settings", open=False), gr.Column(variant='compact', elem_id="txt2img_settings") if shared.opts.img2img_settings_accordion else gr.Column(variant='compact', elem_id="txt2img_settings"): + scripts.scripts_txt2img.prepare_ui() - for category in ordered_ui_categories(): - if category == "prompt": - toprow.create_inline_toprow_prompts() + for category in ordered_ui_categories(): + if category == "prompt": + toprow.create_inline_toprow_prompts() - if category == "sampler": - steps, sampler_name = create_sampler_and_steps_selection(sd_samplers.visible_sampler_names(), "txt2img") + if category == "sampler": + steps, sampler_name = create_sampler_and_steps_selection(sd_samplers.visible_sampler_names(), "txt2img") - elif category == "dimensions": - with FormRow(): - with gr.Column(elem_id="txt2img_column_size", scale=4): - width = gr.Slider(minimum=64, maximum=2048, step=8, label="Width", value=512, elem_id="txt2img_width") - height = gr.Slider(minimum=64, maximum=2048, step=8, label="Height", value=512, elem_id="txt2img_height") + elif category == "dimensions": + with FormRow(): + with gr.Column(elem_id="txt2img_column_size", scale=4): + width = gr.Slider(minimum=64, maximum=2048, step=8, label="Width", value=512, elem_id="txt2img_width") + height = gr.Slider(minimum=64, maximum=2048, step=8, label="Height", value=512, elem_id="txt2img_height") - with gr.Column(elem_id="txt2img_dimensions_row", scale=1, elem_classes="dimensions-tools"): - res_switch_btn = ToolButton(value=switch_values_symbol, elem_id="txt2img_res_switch_btn", tooltip="Switch width/height") + with gr.Column(elem_id="txt2img_dimensions_row", scale=1, elem_classes="dimensions-tools"): + res_switch_btn = ToolButton(value=switch_values_symbol, elem_id="txt2img_res_switch_btn", tooltip="Switch width/height") - if opts.dimensions_and_batch_together: - with gr.Column(elem_id="txt2img_column_batch"): - batch_count = gr.Slider(minimum=1, step=1, label='Batch count', value=1, elem_id="txt2img_batch_count") - batch_size = gr.Slider(minimum=1, maximum=8, step=1, label='Batch size', value=1, elem_id="txt2img_batch_size") - - elif category == "cfg": - with gr.Row(): - cfg_scale = gr.Slider(minimum=1.0, maximum=30.0, step=0.5, label='CFG Scale', value=7.0, elem_id="txt2img_cfg_scale") - - elif category == "checkboxes": - with FormRow(elem_classes="checkboxes-row", variant="compact"): - pass - - elif category == "accordions": - with gr.Row(elem_id="txt2img_accordions", elem_classes="accordions"): - with InputAccordion(False, label="Hires. fix", elem_id="txt2img_hr") as enable_hr: - with enable_hr.extra(): - hr_final_resolution = FormHTML(value="", elem_id="txtimg_hr_finalres", label="Upscaled resolution", interactive=False, min_width=0) - - with FormRow(elem_id="txt2img_hires_fix_row1", variant="compact"): - hr_upscaler = gr.Dropdown(label="Upscaler", elem_id="txt2img_hr_upscaler", choices=[*shared.latent_upscale_modes, *[x.name for x in shared.sd_upscalers]], value=shared.latent_upscale_default_mode) - hr_second_pass_steps = gr.Slider(minimum=0, maximum=150, step=1, label='Hires steps', value=0, elem_id="txt2img_hires_steps") - denoising_strength = gr.Slider(minimum=0.0, maximum=1.0, step=0.01, label='Denoising strength', value=0.7, elem_id="txt2img_denoising_strength") - - with FormRow(elem_id="txt2img_hires_fix_row2", variant="compact"): - hr_scale = gr.Slider(minimum=1.0, maximum=4.0, step=0.05, label="Upscale by", value=2.0, elem_id="txt2img_hr_scale") - hr_resize_x = gr.Slider(minimum=0, maximum=2048, step=8, label="Resize width to", value=0, elem_id="txt2img_hr_resize_x") - hr_resize_y = gr.Slider(minimum=0, maximum=2048, step=8, label="Resize height to", value=0, elem_id="txt2img_hr_resize_y") - - with FormRow(elem_id="txt2img_hires_fix_row3", variant="compact", visible=opts.hires_fix_show_sampler) as hr_sampler_container: - - hr_checkpoint_name = gr.Dropdown(label='Hires checkpoint', elem_id="hr_checkpoint", choices=["Use same checkpoint"] + modules.sd_models.checkpoint_tiles(use_short=True), value="Use same checkpoint") - create_refresh_button(hr_checkpoint_name, modules.sd_models.list_models, lambda: {"choices": ["Use same checkpoint"] + modules.sd_models.checkpoint_tiles(use_short=True)}, "hr_checkpoint_refresh") - - hr_sampler_name = gr.Dropdown(label='Hires sampling method', elem_id="hr_sampler", choices=["Use same sampler"] + sd_samplers.visible_sampler_names(), value="Use same sampler") - - with FormRow(elem_id="txt2img_hires_fix_row4", variant="compact", visible=opts.hires_fix_show_prompts) as hr_prompts_container: - with gr.Column(scale=80): - with gr.Row(): - hr_prompt = gr.Textbox(label="Hires prompt", elem_id="hires_prompt", show_label=False, lines=3, placeholder="Prompt for hires fix pass.\nLeave empty to use the same prompt as in first pass.", elem_classes=["prompt"]) - with gr.Column(scale=80): - with gr.Row(): - hr_negative_prompt = gr.Textbox(label="Hires negative prompt", elem_id="hires_neg_prompt", show_label=False, lines=3, placeholder="Negative prompt for hires fix pass.\nLeave empty to use the same negative prompt as in first pass.", elem_classes=["prompt"]) - - scripts.scripts_txt2img.setup_ui_for_section(category) - - elif category == "batch": - if not opts.dimensions_and_batch_together: - with FormRow(elem_id="txt2img_column_batch"): + if opts.dimensions_and_batch_together: + with gr.Column(elem_id="txt2img_column_batch"): batch_count = gr.Slider(minimum=1, step=1, label='Batch count', value=1, elem_id="txt2img_batch_count") batch_size = gr.Slider(minimum=1, maximum=8, step=1, label='Batch size', value=1, elem_id="txt2img_batch_size") - elif category == "override_settings": - with FormRow(elem_id="txt2img_override_settings_row") as row: - override_settings = create_override_settings_dropdown('txt2img', row) + elif category == "cfg": + with gr.Row(): + cfg_scale = gr.Slider(minimum=1.0, maximum=30.0, step=0.5, label='CFG Scale', value=7.0, elem_id="txt2img_cfg_scale") - elif category == "scripts": - with FormGroup(elem_id="txt2img_script_container"): - custom_inputs = scripts.scripts_txt2img.setup_ui() + elif category == "checkboxes": + with FormRow(elem_classes="checkboxes-row", variant="compact"): + pass + + elif category == "accordions": + with gr.Row(elem_id="txt2img_accordions", elem_classes="accordions"): + with InputAccordion(False, label="Hires. fix", elem_id="txt2img_hr") as enable_hr: + with enable_hr.extra(): + hr_final_resolution = FormHTML(value="", elem_id="txtimg_hr_finalres", label="Upscaled resolution", interactive=False, min_width=0) + + with FormRow(elem_id="txt2img_hires_fix_row1", variant="compact"): + hr_upscaler = gr.Dropdown(label="Upscaler", elem_id="txt2img_hr_upscaler", choices=[*shared.latent_upscale_modes, *[x.name for x in shared.sd_upscalers]], value=shared.latent_upscale_default_mode) + hr_second_pass_steps = gr.Slider(minimum=0, maximum=150, step=1, label='Hires steps', value=0, elem_id="txt2img_hires_steps") + denoising_strength = gr.Slider(minimum=0.0, maximum=1.0, step=0.01, label='Denoising strength', value=0.7, elem_id="txt2img_denoising_strength") + + with FormRow(elem_id="txt2img_hires_fix_row2", variant="compact"): + hr_scale = gr.Slider(minimum=1.0, maximum=4.0, step=0.05, label="Upscale by", value=2.0, elem_id="txt2img_hr_scale") + hr_resize_x = gr.Slider(minimum=0, maximum=2048, step=8, label="Resize width to", value=0, elem_id="txt2img_hr_resize_x") + hr_resize_y = gr.Slider(minimum=0, maximum=2048, step=8, label="Resize height to", value=0, elem_id="txt2img_hr_resize_y") + + with FormRow(elem_id="txt2img_hires_fix_row3", variant="compact", visible=opts.hires_fix_show_sampler) as hr_sampler_container: + + hr_checkpoint_name = gr.Dropdown(label='Hires checkpoint', elem_id="hr_checkpoint", choices=["Use same checkpoint"] + modules.sd_models.checkpoint_tiles(use_short=True), value="Use same checkpoint") + create_refresh_button(hr_checkpoint_name, modules.sd_models.list_models, lambda: {"choices": ["Use same checkpoint"] + modules.sd_models.checkpoint_tiles(use_short=True)}, "hr_checkpoint_refresh") + + hr_sampler_name = gr.Dropdown(label='Hires sampling method', elem_id="hr_sampler", choices=["Use same sampler"] + sd_samplers.visible_sampler_names(), value="Use same sampler") + + with FormRow(elem_id="txt2img_hires_fix_row4", variant="compact", visible=opts.hires_fix_show_prompts) as hr_prompts_container: + with gr.Column(scale=80): + with gr.Row(): + hr_prompt = gr.Textbox(label="Hires prompt", elem_id="hires_prompt", show_label=False, lines=3, placeholder="Prompt for hires fix pass.\nLeave empty to use the same prompt as in first pass.", elem_classes=["prompt"]) + with gr.Column(scale=80): + with gr.Row(): + hr_negative_prompt = gr.Textbox(label="Hires negative prompt", elem_id="hires_neg_prompt", show_label=False, lines=3, placeholder="Negative prompt for hires fix pass.\nLeave empty to use the same negative prompt as in first pass.", elem_classes=["prompt"]) - if category not in {"accordions"}: scripts.scripts_txt2img.setup_ui_for_section(category) + elif category == "batch": + if not opts.dimensions_and_batch_together: + with FormRow(elem_id="txt2img_column_batch"): + batch_count = gr.Slider(minimum=1, step=1, label='Batch count', value=1, elem_id="txt2img_batch_count") + batch_size = gr.Slider(minimum=1, maximum=8, step=1, label='Batch size', value=1, elem_id="txt2img_batch_size") + + elif category == "override_settings": + with FormRow(elem_id="txt2img_override_settings_row") as row: + override_settings = create_override_settings_dropdown('txt2img', row) + + elif category == "scripts": + with FormGroup(elem_id="txt2img_script_container"): + custom_inputs = scripts.scripts_txt2img.setup_ui() + + if category not in {"accordions"}: + scripts.scripts_txt2img.setup_ui_for_section(category) + hr_resolution_preview_inputs = [enable_hr, width, height, hr_scale, hr_resize_x, hr_resize_y] for component in hr_resolution_preview_inputs: @@ -490,258 +489,258 @@ def create_ui(): extra_tabs.__enter__() with gr.Tab("Generation", id="img2img_generation") as img2img_generation_tab, ResizeHandleRow(equal_height=False): - with gr.Accordion("Open for Settings", open=False) if shared.opts.img2img_settings_accordion else gr.Group(): - with gr.Column(variant='compact', elem_id="img2img_settings"): - copy_image_buttons = [] - copy_image_destinations = {} + with gr.Accordion("Open for Settings", open=False), gr.Column(variant='compact', elem_id="img2img_settings") if shared.opts.img2img_settings_accordion else gr.Column(variant='compact', elem_id="img2img_settings"): + copy_image_buttons = [] + copy_image_destinations = {} - def add_copy_image_controls(tab_name, elem): - with gr.Row(variant="compact", elem_id=f"img2img_copy_to_{tab_name}"): - gr.HTML("Copy image to: ", elem_id=f"img2img_label_copy_to_{tab_name}") + def add_copy_image_controls(tab_name, elem): + with gr.Row(variant="compact", elem_id=f"img2img_copy_to_{tab_name}"): + gr.HTML("Copy image to: ", elem_id=f"img2img_label_copy_to_{tab_name}") - for title, name in zip(['img2img', 'sketch', 'inpaint', 'inpaint sketch'], ['img2img', 'sketch', 'inpaint', 'inpaint_sketch']): - if name == tab_name: - gr.Button(title, interactive=False) - copy_image_destinations[name] = elem - continue + for title, name in zip(['img2img', 'sketch', 'inpaint', 'inpaint sketch'], ['img2img', 'sketch', 'inpaint', 'inpaint_sketch']): + if name == tab_name: + gr.Button(title, interactive=False) + copy_image_destinations[name] = elem + continue - button = gr.Button(title) - copy_image_buttons.append((button, name, elem)) + button = gr.Button(title) + copy_image_buttons.append((button, name, elem)) - scripts.scripts_img2img.prepare_ui() + scripts.scripts_img2img.prepare_ui() - for category in ordered_ui_categories(): - if category == "prompt": - toprow.create_inline_toprow_prompts() + for category in ordered_ui_categories(): + if category == "prompt": + toprow.create_inline_toprow_prompts() - if category == "image": - with gr.Tabs(elem_id="mode_img2img"): - img2img_selected_tab = gr.State(0) + if category == "image": + with gr.Tabs(elem_id="mode_img2img"): + img2img_selected_tab = gr.State(0) - with gr.TabItem('img2img', id='img2img', elem_id="img2img_img2img_tab") as tab_img2img: - init_img = gr.Image(label="Image for img2img", elem_id="img2img_image", show_label=False, source="upload", interactive=True, type="pil", tool="editor", image_mode="RGBA", height=opts.img2img_editor_height) - add_copy_image_controls('img2img', init_img) + with gr.TabItem('img2img', id='img2img', elem_id="img2img_img2img_tab") as tab_img2img: + init_img = gr.Image(label="Image for img2img", elem_id="img2img_image", show_label=False, source="upload", interactive=True, type="pil", tool="editor", image_mode="RGBA", height=opts.img2img_editor_height) + add_copy_image_controls('img2img', init_img) - with gr.TabItem('Sketch', id='img2img_sketch', elem_id="img2img_img2img_sketch_tab") as tab_sketch: - sketch = gr.Image(label="Image for img2img", elem_id="img2img_sketch", show_label=False, source="upload", interactive=True, type="pil", tool="color-sketch", image_mode="RGB", height=opts.img2img_editor_height, brush_color=opts.img2img_sketch_default_brush_color) - add_copy_image_controls('sketch', sketch) + with gr.TabItem('Sketch', id='img2img_sketch', elem_id="img2img_img2img_sketch_tab") as tab_sketch: + sketch = gr.Image(label="Image for img2img", elem_id="img2img_sketch", show_label=False, source="upload", interactive=True, type="pil", tool="color-sketch", image_mode="RGB", height=opts.img2img_editor_height, brush_color=opts.img2img_sketch_default_brush_color) + add_copy_image_controls('sketch', sketch) - with gr.TabItem('Inpaint', id='inpaint', elem_id="img2img_inpaint_tab") as tab_inpaint: - init_img_with_mask = gr.Image(label="Image for inpainting with mask", show_label=False, elem_id="img2maskimg", source="upload", interactive=True, type="pil", tool="sketch", image_mode="RGBA", height=opts.img2img_editor_height, brush_color=opts.img2img_inpaint_mask_brush_color) - add_copy_image_controls('inpaint', init_img_with_mask) + with gr.TabItem('Inpaint', id='inpaint', elem_id="img2img_inpaint_tab") as tab_inpaint: + init_img_with_mask = gr.Image(label="Image for inpainting with mask", show_label=False, elem_id="img2maskimg", source="upload", interactive=True, type="pil", tool="sketch", image_mode="RGBA", height=opts.img2img_editor_height, brush_color=opts.img2img_inpaint_mask_brush_color) + add_copy_image_controls('inpaint', init_img_with_mask) - with gr.TabItem('Inpaint sketch', id='inpaint_sketch', elem_id="img2img_inpaint_sketch_tab") as tab_inpaint_color: - inpaint_color_sketch = gr.Image(label="Color sketch inpainting", show_label=False, elem_id="inpaint_sketch", source="upload", interactive=True, type="pil", tool="color-sketch", image_mode="RGB", height=opts.img2img_editor_height, brush_color=opts.img2img_inpaint_sketch_default_brush_color) - inpaint_color_sketch_orig = gr.State(None) - add_copy_image_controls('inpaint_sketch', inpaint_color_sketch) + with gr.TabItem('Inpaint sketch', id='inpaint_sketch', elem_id="img2img_inpaint_sketch_tab") as tab_inpaint_color: + inpaint_color_sketch = gr.Image(label="Color sketch inpainting", show_label=False, elem_id="inpaint_sketch", source="upload", interactive=True, type="pil", tool="color-sketch", image_mode="RGB", height=opts.img2img_editor_height, brush_color=opts.img2img_inpaint_sketch_default_brush_color) + inpaint_color_sketch_orig = gr.State(None) + add_copy_image_controls('inpaint_sketch', inpaint_color_sketch) - def update_orig(image, state): - if image is not None: - same_size = state is not None and state.size == image.size - has_exact_match = np.any(np.all(np.array(image) == np.array(state), axis=-1)) - edited = same_size and has_exact_match - return image if not edited or state is None else state + def update_orig(image, state): + if image is not None: + same_size = state is not None and state.size == image.size + has_exact_match = np.any(np.all(np.array(image) == np.array(state), axis=-1)) + edited = same_size and has_exact_match + return image if not edited or state is None else state - inpaint_color_sketch.change(update_orig, [inpaint_color_sketch, inpaint_color_sketch_orig], inpaint_color_sketch_orig) + inpaint_color_sketch.change(update_orig, [inpaint_color_sketch, inpaint_color_sketch_orig], inpaint_color_sketch_orig) - with gr.TabItem('Inpaint upload', id='inpaint_upload', elem_id="img2img_inpaint_upload_tab") as tab_inpaint_upload: - init_img_inpaint = gr.Image(label="Image for img2img", show_label=False, source="upload", interactive=True, type="pil", elem_id="img_inpaint_base") - init_mask_inpaint = gr.Image(label="Mask", source="upload", interactive=True, type="pil", image_mode="RGBA", elem_id="img_inpaint_mask") + with gr.TabItem('Inpaint upload', id='inpaint_upload', elem_id="img2img_inpaint_upload_tab") as tab_inpaint_upload: + init_img_inpaint = gr.Image(label="Image for img2img", show_label=False, source="upload", interactive=True, type="pil", elem_id="img_inpaint_base") + init_mask_inpaint = gr.Image(label="Mask", source="upload", interactive=True, type="pil", image_mode="RGBA", elem_id="img_inpaint_mask") - with gr.TabItem('Batch', id='batch', elem_id="img2img_batch_tab") as tab_batch: - hidden = '
Disabled when launched with --hide-ui-dir-config.' if shared.cmd_opts.hide_ui_dir_config else '' - gr.HTML( - "

Process images in a directory on the same machine where the server is running." + - "
Use an empty output directory to save pictures normally instead of writing to the output directory." + - f"
Add inpaint batch mask directory to enable inpaint batch processing." - f"{hidden}

" + with gr.TabItem('Batch', id='batch', elem_id="img2img_batch_tab") as tab_batch: + hidden = '
Disabled when launched with --hide-ui-dir-config.' if shared.cmd_opts.hide_ui_dir_config else '' + gr.HTML( + "

Process images in a directory on the same machine where the server is running." + + "
Use an empty output directory to save pictures normally instead of writing to the output directory." + + f"
Add inpaint batch mask directory to enable inpaint batch processing." + f"{hidden}

" + ) + img2img_batch_input_dir = gr.Textbox(label="Input directory", **shared.hide_dirs, elem_id="img2img_batch_input_dir") + img2img_batch_output_dir = gr.Textbox(label="Output directory", **shared.hide_dirs, elem_id="img2img_batch_output_dir") + img2img_batch_inpaint_mask_dir = gr.Textbox(label="Inpaint batch mask directory (required for inpaint batch processing only)", **shared.hide_dirs, elem_id="img2img_batch_inpaint_mask_dir") + with gr.Accordion("PNG info", open=False): + img2img_batch_use_png_info = gr.Checkbox(label="Append png info to prompts", **shared.hide_dirs, elem_id="img2img_batch_use_png_info") + img2img_batch_png_info_dir = gr.Textbox(label="PNG info directory", **shared.hide_dirs, placeholder="Leave empty to use input directory", elem_id="img2img_batch_png_info_dir") + img2img_batch_png_info_props = gr.CheckboxGroup(["Prompt", "Negative prompt", "Seed", "CFG scale", "Sampler", "Steps", "Model hash"], label="Parameters to take from png info", info="Prompts from png info will be appended to prompts set in ui.") + + img2img_tabs = [tab_img2img, tab_sketch, tab_inpaint, tab_inpaint_color, tab_inpaint_upload, tab_batch] + + for i, tab in enumerate(img2img_tabs): + tab.select(fn=lambda tabnum=i: tabnum, inputs=[], outputs=[img2img_selected_tab]) + + def copy_image(img): + if isinstance(img, dict) and 'image' in img: + return img['image'] + + return img + + for button, name, elem in copy_image_buttons: + button.click( + fn=copy_image, + inputs=[elem], + outputs=[copy_image_destinations[name]], + ) + button.click( + fn=lambda: None, + _js=f"switch_to_{name.replace(' ', '_')}", + inputs=[], + outputs=[], + ) + + with FormRow(): + resize_mode = gr.Radio(label="Resize mode", elem_id="resize_mode", choices=["Just resize", "Crop and resize", "Resize and fill", "Just resize (latent upscale)"], type="index", value="Just resize") + + if category == "sampler": + steps, sampler_name = create_sampler_and_steps_selection(sd_samplers.visible_sampler_names(), "img2img") + + elif category == "dimensions": + with FormRow(): + with gr.Column(elem_id="img2img_column_size", scale=4): + selected_scale_tab = gr.State(value=0) + + with gr.Tabs(): + with gr.Tab(label="Resize to", elem_id="img2img_tab_resize_to") as tab_scale_to: + with FormRow(): + with gr.Column(elem_id="img2img_column_size", scale=4): + width = gr.Slider(minimum=64, maximum=2048, step=8, label="Width", value=512, elem_id="img2img_width") + height = gr.Slider(minimum=64, maximum=2048, step=8, label="Height", value=512, elem_id="img2img_height") + with gr.Column(elem_id="img2img_dimensions_row", scale=1, elem_classes="dimensions-tools"): + res_switch_btn = ToolButton(value=switch_values_symbol, elem_id="img2img_res_switch_btn", tooltip="Switch width/height") + detect_image_size_btn = ToolButton(value=detect_image_size_symbol, elem_id="img2img_detect_image_size_btn", tooltip="Auto detect size from img2img") + + with gr.Tab(label="Resize by", elem_id="img2img_tab_resize_by") as tab_scale_by: + scale_by = gr.Slider(minimum=0.05, maximum=4.0, step=0.05, label="Scale", value=1.0, elem_id="img2img_scale") + + with FormRow(): + scale_by_html = FormHTML(resize_from_to_html(0, 0, 0.0), elem_id="img2img_scale_resolution_preview") + gr.Slider(label="Unused", elem_id="img2img_unused_scale_by_slider") + button_update_resize_to = gr.Button(visible=False, elem_id="img2img_update_resize_to") + + on_change_args = dict( + fn=resize_from_to_html, + _js="currentImg2imgSourceResolution", + inputs=[dummy_component, dummy_component, scale_by], + outputs=scale_by_html, + show_progress=False, ) - img2img_batch_input_dir = gr.Textbox(label="Input directory", **shared.hide_dirs, elem_id="img2img_batch_input_dir") - img2img_batch_output_dir = gr.Textbox(label="Output directory", **shared.hide_dirs, elem_id="img2img_batch_output_dir") - img2img_batch_inpaint_mask_dir = gr.Textbox(label="Inpaint batch mask directory (required for inpaint batch processing only)", **shared.hide_dirs, elem_id="img2img_batch_inpaint_mask_dir") - with gr.Accordion("PNG info", open=False): - img2img_batch_use_png_info = gr.Checkbox(label="Append png info to prompts", **shared.hide_dirs, elem_id="img2img_batch_use_png_info") - img2img_batch_png_info_dir = gr.Textbox(label="PNG info directory", **shared.hide_dirs, placeholder="Leave empty to use input directory", elem_id="img2img_batch_png_info_dir") - img2img_batch_png_info_props = gr.CheckboxGroup(["Prompt", "Negative prompt", "Seed", "CFG scale", "Sampler", "Steps", "Model hash"], label="Parameters to take from png info", info="Prompts from png info will be appended to prompts set in ui.") - img2img_tabs = [tab_img2img, tab_sketch, tab_inpaint, tab_inpaint_color, tab_inpaint_upload, tab_batch] + scale_by.release(**on_change_args) + button_update_resize_to.click(**on_change_args) - for i, tab in enumerate(img2img_tabs): - tab.select(fn=lambda tabnum=i: tabnum, inputs=[], outputs=[img2img_selected_tab]) + # the code below is meant to update the resolution label after the image in the image selection UI has changed. + # as it is now the event keeps firing continuously for inpaint edits, which ruins the page with constant requests. + # I assume this must be a gradio bug and for now we'll just do it for non-inpaint inputs. + for component in [init_img, sketch]: + component.change(fn=lambda: None, _js="updateImg2imgResizeToTextAfterChangingImage", inputs=[], outputs=[], show_progress=False) - def copy_image(img): - if isinstance(img, dict) and 'image' in img: - return img['image'] + with FormRow(): + resize_mode = gr.Radio(label="Resize mode", elem_id="resize_mode", choices=["Just resize", "Crop and resize", "Resize and fill", "Just resize (latent upscale)"], type="index", value="Just resize") - return img + scripts.scripts_img2img.prepare_ui() - for button, name, elem in copy_image_buttons: - button.click( - fn=copy_image, - inputs=[elem], - outputs=[copy_image_destinations[name]], - ) - button.click( - fn=lambda: None, - _js=f"switch_to_{name.replace(' ', '_')}", - inputs=[], - outputs=[], - ) + for category in ordered_ui_categories(): + if category == "sampler": + steps, sampler_name = create_sampler_and_steps_selection(sd_samplers.visible_sampler_names(), "img2img") - with FormRow(): - resize_mode = gr.Radio(label="Resize mode", elem_id="resize_mode", choices=["Just resize", "Crop and resize", "Resize and fill", "Just resize (latent upscale)"], type="index", value="Just resize") + elif category == "dimensions": + with FormRow(): + with gr.Column(elem_id="img2img_column_size", scale=4): + selected_scale_tab = gr.State(value=0) - if category == "sampler": - steps, sampler_name = create_sampler_and_steps_selection(sd_samplers.visible_sampler_names(), "img2img") + with gr.Tabs(): + with gr.Tab(label="Resize to", elem_id="img2img_tab_resize_to") as tab_scale_to: + with FormRow(): + with gr.Column(elem_id="img2img_column_size", scale=4): + width = gr.Slider(minimum=64, maximum=2048, step=8, label="Width", value=512, elem_id="img2img_width") + height = gr.Slider(minimum=64, maximum=2048, step=8, label="Height", value=512, elem_id="img2img_height") + with gr.Column(elem_id="img2img_dimensions_row", scale=1, elem_classes="dimensions-tools"): + res_switch_btn = ToolButton(value=switch_values_symbol, elem_id="img2img_res_switch_btn", tooltip="Switch width/height") + detect_image_size_btn = ToolButton(value=detect_image_size_symbol, elem_id="img2img_detect_image_size_btn", tooltip="Auto detect size from img2img") - elif category == "dimensions": - with FormRow(): - with gr.Column(elem_id="img2img_column_size", scale=4): - selected_scale_tab = gr.State(value=0) + with gr.Tab(label="Resize by", elem_id="img2img_tab_resize_by") as tab_scale_by: + scale_by = gr.Slider(minimum=0.05, maximum=4.0, step=0.05, label="Scale", value=1.0, elem_id="img2img_scale") - with gr.Tabs(): - with gr.Tab(label="Resize to", elem_id="img2img_tab_resize_to") as tab_scale_to: - with FormRow(): - with gr.Column(elem_id="img2img_column_size", scale=4): - width = gr.Slider(minimum=64, maximum=2048, step=8, label="Width", value=512, elem_id="img2img_width") - height = gr.Slider(minimum=64, maximum=2048, step=8, label="Height", value=512, elem_id="img2img_height") - with gr.Column(elem_id="img2img_dimensions_row", scale=1, elem_classes="dimensions-tools"): - res_switch_btn = ToolButton(value=switch_values_symbol, elem_id="img2img_res_switch_btn", tooltip="Switch width/height") - detect_image_size_btn = ToolButton(value=detect_image_size_symbol, elem_id="img2img_detect_image_size_btn", tooltip="Auto detect size from img2img") + with FormRow(): + scale_by_html = FormHTML(resize_from_to_html(0, 0, 0.0), elem_id="img2img_scale_resolution_preview") + gr.Slider(label="Unused", elem_id="img2img_unused_scale_by_slider") + button_update_resize_to = gr.Button(visible=False, elem_id="img2img_update_resize_to") - with gr.Tab(label="Resize by", elem_id="img2img_tab_resize_by") as tab_scale_by: - scale_by = gr.Slider(minimum=0.05, maximum=4.0, step=0.05, label="Scale", value=1.0, elem_id="img2img_scale") + on_change_args = dict( + fn=resize_from_to_html, + _js="currentImg2imgSourceResolution", + inputs=[dummy_component, dummy_component, scale_by], + outputs=scale_by_html, + show_progress=False, + ) - with FormRow(): - scale_by_html = FormHTML(resize_from_to_html(0, 0, 0.0), elem_id="img2img_scale_resolution_preview") - gr.Slider(label="Unused", elem_id="img2img_unused_scale_by_slider") - button_update_resize_to = gr.Button(visible=False, elem_id="img2img_update_resize_to") + scale_by.release(**on_change_args) + button_update_resize_to.click(**on_change_args) - on_change_args = dict( - fn=resize_from_to_html, - _js="currentImg2imgSourceResolution", - inputs=[dummy_component, dummy_component, scale_by], - outputs=scale_by_html, - show_progress=False, - ) + # the code below is meant to update the resolution label after the image in the image selection UI has changed. + # as it is now the event keeps firing continuously for inpaint edits, which ruins the page with constant requests. + # I assume this must be a gradio bug and for now we'll just do it for non-inpaint inputs. + for component in [init_img, sketch]: + component.change(fn=lambda: None, _js="updateImg2imgResizeToTextAfterChangingImage", inputs=[], outputs=[], show_progress=False) - scale_by.release(**on_change_args) - button_update_resize_to.click(**on_change_args) + tab_scale_to.select(fn=lambda: 0, inputs=[], outputs=[selected_scale_tab]) + tab_scale_by.select(fn=lambda: 1, inputs=[], outputs=[selected_scale_tab]) - # the code below is meant to update the resolution label after the image in the image selection UI has changed. - # as it is now the event keeps firing continuously for inpaint edits, which ruins the page with constant requests. - # I assume this must be a gradio bug and for now we'll just do it for non-inpaint inputs. - for component in [init_img, sketch]: - component.change(fn=lambda: None, _js="updateImg2imgResizeToTextAfterChangingImage", inputs=[], outputs=[], show_progress=False) - - with FormRow(): - resize_mode = gr.Radio(label="Resize mode", elem_id="resize_mode", choices=["Just resize", "Crop and resize", "Resize and fill", "Just resize (latent upscale)"], type="index", value="Just resize") - - scripts.scripts_img2img.prepare_ui() - - for category in ordered_ui_categories(): - if category == "sampler": - steps, sampler_name = create_sampler_and_steps_selection(sd_samplers.visible_sampler_names(), "img2img") - - elif category == "dimensions": - with FormRow(): - with gr.Column(elem_id="img2img_column_size", scale=4): - selected_scale_tab = gr.State(value=0) - - with gr.Tabs(): - with gr.Tab(label="Resize to", elem_id="img2img_tab_resize_to") as tab_scale_to: - with FormRow(): - with gr.Column(elem_id="img2img_column_size", scale=4): - width = gr.Slider(minimum=64, maximum=2048, step=8, label="Width", value=512, elem_id="img2img_width") - height = gr.Slider(minimum=64, maximum=2048, step=8, label="Height", value=512, elem_id="img2img_height") - with gr.Column(elem_id="img2img_dimensions_row", scale=1, elem_classes="dimensions-tools"): - res_switch_btn = ToolButton(value=switch_values_symbol, elem_id="img2img_res_switch_btn", tooltip="Switch width/height") - detect_image_size_btn = ToolButton(value=detect_image_size_symbol, elem_id="img2img_detect_image_size_btn", tooltip="Auto detect size from img2img") - - with gr.Tab(label="Resize by", elem_id="img2img_tab_resize_by") as tab_scale_by: - scale_by = gr.Slider(minimum=0.05, maximum=4.0, step=0.05, label="Scale", value=1.0, elem_id="img2img_scale") - - with FormRow(): - scale_by_html = FormHTML(resize_from_to_html(0, 0, 0.0), elem_id="img2img_scale_resolution_preview") - gr.Slider(label="Unused", elem_id="img2img_unused_scale_by_slider") - button_update_resize_to = gr.Button(visible=False, elem_id="img2img_update_resize_to") - - on_change_args = dict( - fn=resize_from_to_html, - _js="currentImg2imgSourceResolution", - inputs=[dummy_component, dummy_component, scale_by], - outputs=scale_by_html, - show_progress=False, - ) - - scale_by.release(**on_change_args) - button_update_resize_to.click(**on_change_args) - - # the code below is meant to update the resolution label after the image in the image selection UI has changed. - # as it is now the event keeps firing continuously for inpaint edits, which ruins the page with constant requests. - # I assume this must be a gradio bug and for now we'll just do it for non-inpaint inputs. - for component in [init_img, sketch]: - component.change(fn=lambda: None, _js="updateImg2imgResizeToTextAfterChangingImage", inputs=[], outputs=[], show_progress=False) - - tab_scale_to.select(fn=lambda: 0, inputs=[], outputs=[selected_scale_tab]) - tab_scale_by.select(fn=lambda: 1, inputs=[], outputs=[selected_scale_tab]) - - if opts.dimensions_and_batch_together: - with gr.Column(elem_id="img2img_column_batch"): - batch_count = gr.Slider(minimum=1, step=1, label='Batch count', value=1, elem_id="img2img_batch_count") - batch_size = gr.Slider(minimum=1, maximum=8, step=1, label='Batch size', value=1, elem_id="img2img_batch_size") - - elif category == "denoising": - denoising_strength = gr.Slider(minimum=0.0, maximum=1.0, step=0.01, label='Denoising strength', value=0.75, elem_id="img2img_denoising_strength") - - elif category == "cfg": - with gr.Row(): - cfg_scale = gr.Slider(minimum=1.0, maximum=30.0, step=0.5, label='CFG Scale', value=7.0, elem_id="img2img_cfg_scale") - image_cfg_scale = gr.Slider(minimum=0, maximum=3.0, step=0.05, label='Image CFG Scale', value=1.5, elem_id="img2img_image_cfg_scale", visible=False) - - elif category == "checkboxes": - with FormRow(elem_classes="checkboxes-row", variant="compact"): - pass - - elif category == "accordions": - with gr.Row(elem_id="img2img_accordions", elem_classes="accordions"): - scripts.scripts_img2img.setup_ui_for_section(category) - - elif category == "batch": - if not opts.dimensions_and_batch_together: - with FormRow(elem_id="img2img_column_batch"): + if opts.dimensions_and_batch_together: + with gr.Column(elem_id="img2img_column_batch"): batch_count = gr.Slider(minimum=1, step=1, label='Batch count', value=1, elem_id="img2img_batch_count") batch_size = gr.Slider(minimum=1, maximum=8, step=1, label='Batch size', value=1, elem_id="img2img_batch_size") - elif category == "override_settings": - with FormRow(elem_id="img2img_override_settings_row") as row: - override_settings = create_override_settings_dropdown('img2img', row) + elif category == "denoising": + denoising_strength = gr.Slider(minimum=0.0, maximum=1.0, step=0.01, label='Denoising strength', value=0.75, elem_id="img2img_denoising_strength") - elif category == "scripts": - with FormGroup(elem_id="img2img_script_container"): - custom_inputs = scripts.scripts_img2img.setup_ui() + elif category == "cfg": + with gr.Row(): + cfg_scale = gr.Slider(minimum=1.0, maximum=30.0, step=0.5, label='CFG Scale', value=7.0, elem_id="img2img_cfg_scale") + image_cfg_scale = gr.Slider(minimum=0, maximum=3.0, step=0.05, label='Image CFG Scale', value=1.5, elem_id="img2img_image_cfg_scale", visible=False) - elif category == "inpaint": - with FormGroup(elem_id="inpaint_controls", visible=False) as inpaint_controls: - with FormRow(): - mask_blur = gr.Slider(label='Mask blur', minimum=0, maximum=64, step=1, value=4, elem_id="img2img_mask_blur") - mask_alpha = gr.Slider(label="Mask transparency", visible=False, elem_id="img2img_mask_alpha") + elif category == "checkboxes": + with FormRow(elem_classes="checkboxes-row", variant="compact"): + pass - with FormRow(): - inpainting_mask_invert = gr.Radio(label='Mask mode', choices=['Inpaint masked', 'Inpaint not masked'], value='Inpaint masked', type="index", elem_id="img2img_mask_mode") - - with FormRow(): - inpainting_fill = gr.Radio(label='Masked content', choices=['fill', 'original', 'latent noise', 'latent nothing'], value='original', type="index", elem_id="img2img_inpainting_fill") - - with FormRow(): - with gr.Column(): - inpaint_full_res = gr.Radio(label="Inpaint area", choices=["Whole picture", "Only masked"], type="index", value="Whole picture", elem_id="img2img_inpaint_full_res") - - with gr.Column(scale=4): - inpaint_full_res_padding = gr.Slider(label='Only masked padding, pixels', minimum=0, maximum=256, step=4, value=32, elem_id="img2img_inpaint_full_res_padding") - - if category not in {"accordions"}: + elif category == "accordions": + with gr.Row(elem_id="img2img_accordions", elem_classes="accordions"): scripts.scripts_img2img.setup_ui_for_section(category) + + elif category == "batch": + if not opts.dimensions_and_batch_together: + with FormRow(elem_id="img2img_column_batch"): + batch_count = gr.Slider(minimum=1, step=1, label='Batch count', value=1, elem_id="img2img_batch_count") + batch_size = gr.Slider(minimum=1, maximum=8, step=1, label='Batch size', value=1, elem_id="img2img_batch_size") + + elif category == "override_settings": + with FormRow(elem_id="img2img_override_settings_row") as row: + override_settings = create_override_settings_dropdown('img2img', row) + + elif category == "scripts": + with FormGroup(elem_id="img2img_script_container"): + custom_inputs = scripts.scripts_img2img.setup_ui() + + elif category == "inpaint": + with FormGroup(elem_id="inpaint_controls", visible=False) as inpaint_controls: + with FormRow(): + mask_blur = gr.Slider(label='Mask blur', minimum=0, maximum=64, step=1, value=4, elem_id="img2img_mask_blur") + mask_alpha = gr.Slider(label="Mask transparency", visible=False, elem_id="img2img_mask_alpha") + + with FormRow(): + inpainting_mask_invert = gr.Radio(label='Mask mode', choices=['Inpaint masked', 'Inpaint not masked'], value='Inpaint masked', type="index", elem_id="img2img_mask_mode") + + with FormRow(): + inpainting_fill = gr.Radio(label='Masked content', choices=['fill', 'original', 'latent noise', 'latent nothing'], value='original', type="index", elem_id="img2img_inpainting_fill") + + with FormRow(): + with gr.Column(): + inpaint_full_res = gr.Radio(label="Inpaint area", choices=["Whole picture", "Only masked"], type="index", value="Whole picture", elem_id="img2img_inpaint_full_res") + + with gr.Column(scale=4): + inpaint_full_res_padding = gr.Slider(label='Only masked padding, pixels', minimum=0, maximum=256, step=4, value=32, elem_id="img2img_inpaint_full_res_padding") + + if category not in {"accordions"}: + scripts.scripts_img2img.setup_ui_for_section(category) + def select_img2img_tab(tab): return gr.update(visible=tab in [2, 3, 4]), gr.update(visible=tab == 3), From 9aa4d098f07655d99cd16e8e9c984d043dbf9006 Mon Sep 17 00:00:00 2001 From: Emily Zeng Date: Thu, 9 Nov 2023 13:25:24 -0500 Subject: [PATCH 030/139] removed changes that weren't merged properly --- modules/ui.py | 51 +-------------------------------------------------- 1 file changed, 1 insertion(+), 50 deletions(-) diff --git a/modules/ui.py b/modules/ui.py index 0faccbd34..3eec7839f 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -592,55 +592,6 @@ def create_ui(): if category == "sampler": steps, sampler_name = create_sampler_and_steps_selection(sd_samplers.visible_sampler_names(), "img2img") - elif category == "dimensions": - with FormRow(): - with gr.Column(elem_id="img2img_column_size", scale=4): - selected_scale_tab = gr.State(value=0) - - with gr.Tabs(): - with gr.Tab(label="Resize to", elem_id="img2img_tab_resize_to") as tab_scale_to: - with FormRow(): - with gr.Column(elem_id="img2img_column_size", scale=4): - width = gr.Slider(minimum=64, maximum=2048, step=8, label="Width", value=512, elem_id="img2img_width") - height = gr.Slider(minimum=64, maximum=2048, step=8, label="Height", value=512, elem_id="img2img_height") - with gr.Column(elem_id="img2img_dimensions_row", scale=1, elem_classes="dimensions-tools"): - res_switch_btn = ToolButton(value=switch_values_symbol, elem_id="img2img_res_switch_btn", tooltip="Switch width/height") - detect_image_size_btn = ToolButton(value=detect_image_size_symbol, elem_id="img2img_detect_image_size_btn", tooltip="Auto detect size from img2img") - - with gr.Tab(label="Resize by", elem_id="img2img_tab_resize_by") as tab_scale_by: - scale_by = gr.Slider(minimum=0.05, maximum=4.0, step=0.05, label="Scale", value=1.0, elem_id="img2img_scale") - - with FormRow(): - scale_by_html = FormHTML(resize_from_to_html(0, 0, 0.0), elem_id="img2img_scale_resolution_preview") - gr.Slider(label="Unused", elem_id="img2img_unused_scale_by_slider") - button_update_resize_to = gr.Button(visible=False, elem_id="img2img_update_resize_to") - - on_change_args = dict( - fn=resize_from_to_html, - _js="currentImg2imgSourceResolution", - inputs=[dummy_component, dummy_component, scale_by], - outputs=scale_by_html, - show_progress=False, - ) - - scale_by.release(**on_change_args) - button_update_resize_to.click(**on_change_args) - - # the code below is meant to update the resolution label after the image in the image selection UI has changed. - # as it is now the event keeps firing continuously for inpaint edits, which ruins the page with constant requests. - # I assume this must be a gradio bug and for now we'll just do it for non-inpaint inputs. - for component in [init_img, sketch]: - component.change(fn=lambda: None, _js="updateImg2imgResizeToTextAfterChangingImage", inputs=[], outputs=[], show_progress=False) - - with FormRow(): - resize_mode = gr.Radio(label="Resize mode", elem_id="resize_mode", choices=["Just resize", "Crop and resize", "Resize and fill", "Just resize (latent upscale)"], type="index", value="Just resize") - - scripts.scripts_img2img.prepare_ui() - - for category in ordered_ui_categories(): - if category == "sampler": - steps, sampler_name = create_sampler_and_steps_selection(sd_samplers.visible_sampler_names(), "img2img") - elif category == "dimensions": with FormRow(): with gr.Column(elem_id="img2img_column_size", scale=4): @@ -740,7 +691,7 @@ def create_ui(): if category not in {"accordions"}: scripts.scripts_img2img.setup_ui_for_section(category) - + def select_img2img_tab(tab): return gr.update(visible=tab in [2, 3, 4]), gr.update(visible=tab == 3), From ff2952f10551aab2000002079d5f862af979e964 Mon Sep 17 00:00:00 2001 From: Emily Zeng Date: Thu, 9 Nov 2023 13:35:52 -0500 Subject: [PATCH 031/139] multiline with statement for readibility --- modules/ui.py | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-) diff --git a/modules/ui.py b/modules/ui.py index 3eec7839f..bf06776e9 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -270,7 +270,9 @@ def create_ui(): extra_tabs.__enter__() with gr.Tab("Generation", id="txt2img_generation") as txt2img_generation_tab, ResizeHandleRow(equal_height=False): - with gr.Accordion("Open for Settings", open=False), gr.Column(variant='compact', elem_id="txt2img_settings") if shared.opts.img2img_settings_accordion else gr.Column(variant='compact', elem_id="txt2img_settings"): + with gr.Accordion("Open for Settings", open=False), gr.Column(variant='compact', elem_id="txt2img_settings") \ + if shared.opts.img2img_settings_accordion else gr.Column(variant='compact', elem_id="txt2img_settings"): + scripts.scripts_txt2img.prepare_ui() for category in ordered_ui_categories(): @@ -489,7 +491,9 @@ def create_ui(): extra_tabs.__enter__() with gr.Tab("Generation", id="img2img_generation") as img2img_generation_tab, ResizeHandleRow(equal_height=False): - with gr.Accordion("Open for Settings", open=False), gr.Column(variant='compact', elem_id="img2img_settings") if shared.opts.img2img_settings_accordion else gr.Column(variant='compact', elem_id="img2img_settings"): + with gr.Accordion("Open for Settings", open=False), gr.Column(variant='compact', elem_id="img2img_settings") \ + if shared.opts.img2img_settings_accordion else gr.Column(variant='compact', elem_id="img2img_settings"): + copy_image_buttons = [] copy_image_destinations = {} From 6d77a6e1c6b27ae82b2186cfc36cc4ad2a5e9ecf Mon Sep 17 00:00:00 2001 From: "fuchen.ljl" Date: Fri, 10 Nov 2023 14:40:39 +0800 Subject: [PATCH 033/139] Update README.md Modify the stablediffusion dependency address --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index 4e0834400..1c97ecbb0 100644 --- a/README.md +++ b/README.md @@ -146,7 +146,7 @@ For the purposes of getting Google and other search engines to crawl the wiki, h ## Credits Licenses for borrowed code can be found in `Settings -> Licenses` screen, and also in `html/licenses.html` file. -- Stable Diffusion - https://github.com/CompVis/stable-diffusion, https://github.com/CompVis/taming-transformers +- Stable Diffusion - https://github.com/Stability-AI/stablediffusion, https://github.com/CompVis/taming-transformers - k-diffusion - https://github.com/crowsonkb/k-diffusion.git - GFPGAN - https://github.com/TencentARC/GFPGAN.git - CodeFormer - https://github.com/sczhou/CodeFormer From 66767e3876dde8d0ef27ce00254cd6b75332f036 Mon Sep 17 00:00:00 2001 From: "Alessandro de Oliveira Faria (A.K.A. CABELO)" Date: Fri, 10 Nov 2023 03:45:44 -0300 Subject: [PATCH 034/139] - opensuse compatibility --- README.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/README.md b/README.md index 4e0834400..89e54a613 100644 --- a/README.md +++ b/README.md @@ -121,6 +121,8 @@ Alternatively, use online services (like Google Colab): sudo apt install wget git python3 python3-venv libgl1 libglib2.0-0 # Red Hat-based: sudo dnf install wget git python3 +# openSUSE-based: +sudo zypper install wget git python3 # Arch-based: sudo pacman -S wget git python3 ``` From 7ff54005fee46ce188544db75c27de61ae279001 Mon Sep 17 00:00:00 2001 From: missionfloyd Date: Thu, 9 Nov 2023 23:47:53 -0700 Subject: [PATCH 035/139] Enable prompt hotkeys in style editor --- modules/ui_prompt_styles.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/modules/ui_prompt_styles.py b/modules/ui_prompt_styles.py index 85eb3a641..d6f8d4c76 100644 --- a/modules/ui_prompt_styles.py +++ b/modules/ui_prompt_styles.py @@ -64,10 +64,10 @@ class UiPromptStyles: self.materialize = ui_components.ToolButton(value=styles_materialize_symbol, elem_id=f"{tabname}_style_apply", tooltip="Apply all selected styles from the style selction dropdown in main UI to the prompt.") with gr.Row(): - self.prompt = gr.Textbox(label="Prompt", show_label=True, elem_id=f"{tabname}_edit_style_prompt", lines=3) + self.prompt = gr.Textbox(label="Prompt", show_label=True, elem_id=f"{tabname}_edit_style_prompt", lines=3, elem_classes=["prompt"]) with gr.Row(): - self.neg_prompt = gr.Textbox(label="Negative prompt", show_label=True, elem_id=f"{tabname}_edit_style_neg_prompt", lines=3) + self.neg_prompt = gr.Textbox(label="Negative prompt", show_label=True, elem_id=f"{tabname}_edit_style_neg_prompt", lines=3, elem_classes=["prompt"]) with gr.Row(): self.save = gr.Button('Save', variant='primary', elem_id=f'{tabname}_edit_style_save', visible=False) From 6a86b3ad9bc7bb9a58dc4228ecf93a3a511ed122 Mon Sep 17 00:00:00 2001 From: "Alessandro de Oliveira Faria (A.K.A. CABELO)" Date: Fri, 10 Nov 2023 14:15:34 -0300 Subject: [PATCH 036/139] Compatibility with Debian 11, Fedora 34+ and openSUSE 15.4+ --- README.md | 4 ++-- webui.sh | 4 ++-- 2 files changed, 4 insertions(+), 4 deletions(-) diff --git a/README.md b/README.md index 89e54a613..d4aa376b1 100644 --- a/README.md +++ b/README.md @@ -120,9 +120,9 @@ Alternatively, use online services (like Google Colab): # Debian-based: sudo apt install wget git python3 python3-venv libgl1 libglib2.0-0 # Red Hat-based: -sudo dnf install wget git python3 +sudo dnf install wget git python3 gperftools-libs libglvnd-glx # openSUSE-based: -sudo zypper install wget git python3 +sudo zypper install wget git python3 libtcmalloc4 libglvnd # Arch-based: sudo pacman -S wget git python3 ``` diff --git a/webui.sh b/webui.sh index 3d0f87eed..5c23c1d8f 100755 --- a/webui.sh +++ b/webui.sh @@ -87,7 +87,7 @@ delimiter="################################################################" printf "\n%s\n" "${delimiter}" printf "\e[1m\e[32mInstall script for stable-diffusion + Web UI\n" -printf "\e[1m\e[34mTested on Debian 11 (Bullseye)\e[0m" +printf "\e[1m\e[34mTested on Debian 11 (Bullseye), Fedora 34+ and openSUSE Leap 15.4 or newer.\e[0m" printf "\n%s\n" "${delimiter}" # Do not run as root @@ -222,7 +222,7 @@ fi # Try using TCMalloc on Linux prepare_tcmalloc() { if [[ "${OSTYPE}" == "linux"* ]] && [[ -z "${NO_TCMALLOC}" ]] && [[ -z "${LD_PRELOAD}" ]]; then - TCMALLOC="$(PATH=/usr/sbin:$PATH ldconfig -p | grep -Po "libtcmalloc(_minimal|)\.so\.\d" | head -n 1)" + TCMALLOC="$(PATH=/sbin:$PATH ldconfig -p | grep -Po "libtcmalloc(_minimal|)\.so\.\d" | head -n 1)" if [[ ! -z "${TCMALLOC}" ]]; then echo "Using TCMalloc: ${TCMALLOC}" export LD_PRELOAD="${TCMALLOC}" From 5432d9301359945b595d5e6649c7a64b4bb0bfca Mon Sep 17 00:00:00 2001 From: w-e-w <40751091+w-e-w@users.noreply.github.com> Date: Sat, 11 Nov 2023 03:38:55 +0900 Subject: [PATCH 037/139] fix added accordion settings options --- modules/ui.py | 8 +++----- 1 file changed, 3 insertions(+), 5 deletions(-) diff --git a/modules/ui.py b/modules/ui.py index bf06776e9..f28de3543 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -4,6 +4,7 @@ import os import sys from functools import reduce import warnings +from contextlib import suppress import gradio as gr import gradio.utils @@ -270,9 +271,7 @@ def create_ui(): extra_tabs.__enter__() with gr.Tab("Generation", id="txt2img_generation") as txt2img_generation_tab, ResizeHandleRow(equal_height=False): - with gr.Accordion("Open for Settings", open=False), gr.Column(variant='compact', elem_id="txt2img_settings") \ - if shared.opts.img2img_settings_accordion else gr.Column(variant='compact', elem_id="txt2img_settings"): - + with gr.Accordion("Open for Settings", open=False) if shared.opts.txt2img_settings_accordion else suppress(), gr.Column(variant='compact', elem_id="txt2img_settings"): scripts.scripts_txt2img.prepare_ui() for category in ordered_ui_categories(): @@ -491,8 +490,7 @@ def create_ui(): extra_tabs.__enter__() with gr.Tab("Generation", id="img2img_generation") as img2img_generation_tab, ResizeHandleRow(equal_height=False): - with gr.Accordion("Open for Settings", open=False), gr.Column(variant='compact', elem_id="img2img_settings") \ - if shared.opts.img2img_settings_accordion else gr.Column(variant='compact', elem_id="img2img_settings"): + with gr.Accordion("Open for Settings", open=False) if shared.opts.img2img_settings_accordion else suppress(), gr.Column(variant='compact', elem_id="img2img_settings"): copy_image_buttons = [] copy_image_destinations = {} From 3a4a6c43a4ca31056d5c09bb54e3eef24e6cf864 Mon Sep 17 00:00:00 2001 From: Emily Zeng Date: Fri, 10 Nov 2023 16:06:01 -0500 Subject: [PATCH 038/139] ExitStack as alternative to suppress --- modules/ui.py | 13 ++++++++++--- 1 file changed, 10 insertions(+), 3 deletions(-) diff --git a/modules/ui.py b/modules/ui.py index f28de3543..ba0d8542b 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -4,7 +4,7 @@ import os import sys from functools import reduce import warnings -from contextlib import suppress +from contextlib import ExitStack import gradio as gr import gradio.utils @@ -271,7 +271,11 @@ def create_ui(): extra_tabs.__enter__() with gr.Tab("Generation", id="txt2img_generation") as txt2img_generation_tab, ResizeHandleRow(equal_height=False): - with gr.Accordion("Open for Settings", open=False) if shared.opts.txt2img_settings_accordion else suppress(), gr.Column(variant='compact', elem_id="txt2img_settings"): + with ExitStack() as stack: + if shared.opts.txt2img_settings_accordion: + stack.enter_context(gr.Accordion("Open for Settings", open=False)) + stack.enter_context(gr.Column(variant='compact', elem_id="txt2img_settings")) + scripts.scripts_txt2img.prepare_ui() for category in ordered_ui_categories(): @@ -490,7 +494,10 @@ def create_ui(): extra_tabs.__enter__() with gr.Tab("Generation", id="img2img_generation") as img2img_generation_tab, ResizeHandleRow(equal_height=False): - with gr.Accordion("Open for Settings", open=False) if shared.opts.img2img_settings_accordion else suppress(), gr.Column(variant='compact', elem_id="img2img_settings"): + with ExitStack() as stack: + if shared.opts.img2img_settings_accordion: + stack.enter_context(gr.Accordion("Open for Settings", open=False)) + stack.enter_context(gr.Column(variant='compact', elem_id="img2img_settings")) copy_image_buttons = [] copy_image_destinations = {} From 0fc7dc1c04a046d95588651ffc4e71a7d40378d3 Mon Sep 17 00:00:00 2001 From: wfjsw Date: Sat, 11 Nov 2023 04:01:13 -0600 Subject: [PATCH 039/139] implementing script metadata and DAG sorting mechanism --- modules/extensions.py | 80 +++++++++++++++++++++--- modules/scripts.py | 139 +++++++++++++++++++++++++++++++++++++----- 2 files changed, 196 insertions(+), 23 deletions(-) diff --git a/modules/extensions.py b/modules/extensions.py index bf9a1878f..e317a4041 100644 --- a/modules/extensions.py +++ b/modules/extensions.py @@ -1,3 +1,5 @@ +import configparser +import functools import os import threading @@ -23,8 +25,9 @@ class Extension: lock = threading.Lock() cached_fields = ['remote', 'commit_date', 'branch', 'commit_hash', 'version'] - def __init__(self, name, path, enabled=True, is_builtin=False): + def __init__(self, name, path, enabled=True, is_builtin=False, canonical_name=None): self.name = name + self.canonical_name = canonical_name or name.lower() self.path = path self.enabled = enabled self.status = '' @@ -37,6 +40,17 @@ class Extension: self.remote = None self.have_info_from_repo = False + @functools.cached_property + def metadata(self): + if os.path.isfile(os.path.join(self.path, "sd_webui_metadata.ini")): + try: + config = configparser.ConfigParser() + config.read(os.path.join(self.path, "sd_webui_metadata.ini")) + return config + except Exception: + errors.report(f"Error reading sd_webui_metadata.ini for extension {self.canonical_name}.", exc_info=True) + return None + def to_dict(self): return {x: getattr(self, x) for x in self.cached_fields} @@ -136,9 +150,6 @@ class Extension: def list_extensions(): extensions.clear() - if not os.path.isdir(extensions_dir): - return - if shared.cmd_opts.disable_all_extensions: print("*** \"--disable-all-extensions\" arg was used, will not load any extensions ***") elif shared.opts.disable_all_extensions == "all": @@ -148,18 +159,69 @@ def list_extensions(): elif shared.opts.disable_all_extensions == "extra": print("*** \"Disable all extensions\" option was set, will only load built-in extensions ***") - extension_paths = [] + extension_dependency_map = {} + + # scan through extensions directory and load metadata for dirname in [extensions_dir, extensions_builtin_dir]: if not os.path.isdir(dirname): - return + continue for extension_dirname in sorted(os.listdir(dirname)): path = os.path.join(dirname, extension_dirname) if not os.path.isdir(path): continue - extension_paths.append((extension_dirname, path, dirname == extensions_builtin_dir)) + canonical_name = extension_dirname + requires = None - for dirname, path, is_builtin in extension_paths: - extension = Extension(name=dirname, path=path, enabled=dirname not in shared.opts.disabled_extensions, is_builtin=is_builtin) + if os.path.isfile(os.path.join(path, "sd_webui_metadata.ini")): + try: + config = configparser.ConfigParser() + config.read(os.path.join(path, "sd_webui_metadata.ini")) + canonical_name = config.get("Extension", "Name", fallback=canonical_name) + requires = config.get("Extension", "Requires", fallback=None) + continue + except Exception: + errors.report(f"Error reading sd_webui_metadata.ini for extension {extension_dirname}. " + f"Will load regardless.", exc_info=True) + + canonical_name = canonical_name.lower().strip() + + # check for duplicated canonical names + if canonical_name in extension_dependency_map: + errors.report(f"Duplicate canonical name \"{canonical_name}\" found in extensions " + f"\"{extension_dirname}\" and \"{extension_dependency_map[canonical_name]['dirname']}\". " + f"The current loading extension will be discarded.", exc_info=False) + continue + + # we want to wash the data to lowercase and remove whitespaces just in case + requires = [x.strip() for x in requires.lower().split(',')] if requires else [] + + extension_dependency_map[canonical_name] = { + "dirname": extension_dirname, + "path": path, + "requires": requires, + } + + # check for requirements + for (_, extension_data) in extension_dependency_map.items(): + dirname, path, requires = extension_data['dirname'], extension_data['path'], extension_data['requires'] + requirement_met = True + for req in requires: + if req not in extension_dependency_map: + errors.report(f"Extension \"{dirname}\" requires \"{req}\" which is not installed. " + f"The current loading extension will be discarded.", exc_info=False) + requirement_met = False + break + dep_dirname = extension_dependency_map[req]['dirname'] + if dep_dirname in shared.opts.disabled_extensions: + errors.report(f"Extension \"{dirname}\" requires \"{dep_dirname}\" which is disabled. " + f"The current loading extension will be discarded.", exc_info=False) + requirement_met = False + break + + is_builtin = dirname == extensions_builtin_dir + extension = Extension(name=dirname, path=path, + enabled=dirname not in shared.opts.disabled_extensions and requirement_met, + is_builtin=is_builtin) extensions.append(extension) diff --git a/modules/scripts.py b/modules/scripts.py index 5c6e0226e..e92a34a0e 100644 --- a/modules/scripts.py +++ b/modules/scripts.py @@ -2,6 +2,7 @@ import os import re import sys import inspect +from graphlib import TopologicalSorter, CycleError from collections import namedtuple from dataclasses import dataclass @@ -314,15 +315,131 @@ ScriptClassData = namedtuple("ScriptClassData", ["script_class", "path", "basedi def list_scripts(scriptdirname, extension, *, include_extensions=True): scripts_list = [] + script_dependency_map = {} - basedir = os.path.join(paths.script_path, scriptdirname) - if os.path.exists(basedir): - for filename in sorted(os.listdir(basedir)): - scripts_list.append(ScriptFile(paths.script_path, filename, os.path.join(basedir, filename))) + # build script dependency map + + root_script_basedir = os.path.join(paths.script_path, scriptdirname) + if os.path.exists(root_script_basedir): + for filename in sorted(os.listdir(root_script_basedir)): + script_dependency_map[filename] = { + "extension": None, + "extension_dirname": None, + "script_file": ScriptFile(paths.script_path, filename, os.path.join(root_script_basedir, filename)), + "requires": [], + "load_before": [], + "load_after": [], + } if include_extensions: for ext in extensions.active(): - scripts_list += ext.list_files(scriptdirname, extension) + extension_scripts_list = ext.list_files(scriptdirname, extension) + for extension_script in extension_scripts_list: + # this is built on the assumption that script name is unique. + # I think bad thing is gonna happen if name collide in the current implementation anyway, but we + # will need to refactor here if this assumption is broken later on. + if extension_script.filename in script_dependency_map: + errors.report(f"Duplicate script name \"{extension_script.filename}\" found in extensions " + f"\"{ext.name}\" and \"{script_dependency_map[extension_script.filename]['extension_dirname'] or 'builtin'}\". " + f"The current loading file will be discarded.", exc_info=False) + continue + + relative_path = scriptdirname + "/" + extension_script.filename + + requires = None + load_before = None + load_after = None + + if ext.metadata is not None: + requires = ext.metadata.get(relative_path, "Requires", fallback=None) + load_before = ext.metadata.get(relative_path, "Before", fallback=None) + load_after = ext.metadata.get(relative_path, "After", fallback=None) + + requires = [x.strip() for x in requires.split(',')] if requires else [] + load_after = [x.strip() for x in load_after.split(',')] if load_after else [] + load_before = [x.strip() for x in load_before.split(',')] if load_before else [] + + script_dependency_map[extension_script.filename] = { + "extension": ext.canonical_name, + "extension_dirname": ext.name, + "script_file": extension_script, + "requires": requires, + "load_before": load_before, + "load_after": load_after, + } + + # resolve dependencies + + loaded_extensions = set() + for _, script_data in script_dependency_map.items(): + if script_data['extension'] is not None: + loaded_extensions.add(script_data['extension']) + + for script_filename, script_data in script_dependency_map.items(): + # load before requires inverse dependency + # in this case, append the script name into the load_after list of the specified script + for load_before_script in script_data['load_before']: + if load_before_script.startswith('ext:'): + # if this requires an extension to be loaded before + required_extension = load_before_script[4:] + for _, script_data2 in script_dependency_map.items(): + if script_data2['extension'] == required_extension: + script_data2['load_after'].append(script_filename) + break + else: + # if this requires an individual script to be loaded before + if load_before_script in script_dependency_map: + script_dependency_map[load_before_script]['load_after'].append(script_filename) + + # resolve extension name in load_after lists + for load_after_script in script_data['load_after']: + if load_after_script.startswith('ext:'): + # if this requires an extension to be loaded after + required_extension = load_after_script[4:] + for script_file_name2, script_data2 in script_dependency_map.items(): + if script_data2['extension'] == required_extension: + script_data['load_after'].append(script_file_name2) + + # remove all extension names in load_after lists + script_data['load_after'] = [x for x in script_data['load_after'] if not x.startswith('ext:')] + + # build the DAG + sorter = TopologicalSorter() + for script_filename, script_data in script_dependency_map.items(): + requirement_met = True + for required_script in script_data['requires']: + if required_script.startswith('ext:'): + # if this requires an extension to be installed + required_extension = required_script[4:] + if required_extension not in loaded_extensions: + errors.report(f"Script \"{script_filename}\" requires extension \"{required_extension}\" to " + f"be loaded, but it is not. Skipping.", + exc_info=False) + requirement_met = False + break + else: + # if this requires an individual script to be loaded + if required_script not in script_dependency_map: + errors.report(f"Script \"{script_filename}\" requires script \"{required_script}\" to " + f"be loaded, but it is not. Skipping.", + exc_info=False) + requirement_met = False + break + if not requirement_met: + continue + + sorter.add(script_filename, *script_data['load_after']) + + # sort the scripts + try: + ordered_script = sorter.static_order() + except CycleError: + errors.report("Cycle detected in script dependencies. Scripts will load in ascending order.", exc_info=True) + ordered_script = script_dependency_map.keys() + + for script_filename in ordered_script: + script_data = script_dependency_map[script_filename] + scripts_list.append(script_data['script_file']) scripts_list = [x for x in scripts_list if os.path.splitext(x.path)[1].lower() == extension and os.path.isfile(x.path)] @@ -365,15 +482,9 @@ def load_scripts(): elif issubclass(script_class, scripts_postprocessing.ScriptPostprocessing): postprocessing_scripts_data.append(ScriptClassData(script_class, scriptfile.path, scriptfile.basedir, module)) - def orderby(basedir): - # 1st webui, 2nd extensions-builtin, 3rd extensions - priority = {os.path.join(paths.script_path, "extensions-builtin"):1, paths.script_path:0} - for key in priority: - if basedir.startswith(key): - return priority[key] - return 9999 - - for scriptfile in sorted(scripts_list, key=lambda x: [orderby(x.basedir), x]): + # here the scripts_list is already ordered + # processing_script is not considered though + for scriptfile in scripts_list: try: if scriptfile.basedir != paths.script_path: sys.path = [scriptfile.basedir] + sys.path From 0d1924c48be3d02650e87b12a4f53165a8b4a599 Mon Sep 17 00:00:00 2001 From: wfjsw Date: Sat, 11 Nov 2023 04:03:55 -0600 Subject: [PATCH 040/139] populate loaded_extensions from extension list instead --- modules/scripts.py | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) diff --git a/modules/scripts.py b/modules/scripts.py index e92a34a0e..7cdf288df 100644 --- a/modules/scripts.py +++ b/modules/scripts.py @@ -371,9 +371,8 @@ def list_scripts(scriptdirname, extension, *, include_extensions=True): # resolve dependencies loaded_extensions = set() - for _, script_data in script_dependency_map.items(): - if script_data['extension'] is not None: - loaded_extensions.add(script_data['extension']) + for ext in extensions.active(): + loaded_extensions.add(ext.canonical_name) for script_filename, script_data in script_dependency_map.items(): # load before requires inverse dependency From bc1a450124ab643fc0c3ea7630d875afb4b84b84 Mon Sep 17 00:00:00 2001 From: wfjsw Date: Sat, 11 Nov 2023 04:08:45 -0600 Subject: [PATCH 041/139] reverse the extension load order so builtin extensions load earlier natively --- modules/extensions.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/modules/extensions.py b/modules/extensions.py index e317a4041..7583a3b03 100644 --- a/modules/extensions.py +++ b/modules/extensions.py @@ -162,7 +162,7 @@ def list_extensions(): extension_dependency_map = {} # scan through extensions directory and load metadata - for dirname in [extensions_dir, extensions_builtin_dir]: + for dirname in [extensions_builtin_dir, extensions_dir]: if not os.path.isdir(dirname): continue From 294f8a514f982248cda1cafda30d35566f3a0321 Mon Sep 17 00:00:00 2001 From: aria1th <35677394+aria1th@users.noreply.github.com> Date: Sat, 11 Nov 2023 23:28:12 +0900 Subject: [PATCH 042/139] add hyperTile https://github.com/tfernd/HyperTile --- modules/processing.py | 27 ++++++++++++++++++++++++--- modules/shared_options.py | 2 ++ 2 files changed, 26 insertions(+), 3 deletions(-) diff --git a/modules/processing.py b/modules/processing.py index b0e240a46..e23095343 100644 --- a/modules/processing.py +++ b/modules/processing.py @@ -799,6 +799,16 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed: infotexts = [] output_images = [] + unet_object = p.sd_model.model + vae_model = p.sd_model.first_stage_model + try: + from hyper_tile import split_attention, flush + except (ImportError, ModuleNotFoundError): # pip install git+https://github.com/tfernd/HyperTile@2ef64b2800d007d305755c33550537410310d7df + split_attention = lambda *args, **kwargs: lambda x: x # return a no-op context manager + flush = lambda: None + import random + saved_rng_state = random.getstate() + random.seed(p.seed) # hyper_tile uses random, so we need to seed it with torch.no_grad(), p.sd_model.ema_scope(): with devices.autocast(): @@ -866,15 +876,25 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed: shared.state.job = f"Batch {n+1} out of {p.n_iter}" with devices.without_autocast() if devices.unet_needs_upcast else devices.autocast(): - samples_ddim = p.sample(conditioning=p.c, unconditional_conditioning=p.uc, seeds=p.seeds, subseeds=p.subseeds, subseed_strength=p.subseed_strength, prompts=p.prompts) + # get largest tile size available, which is 2^x which is factor of gcd of p.width and p.height + gcd = math.gcd(p.width, p.height) + largest_tile_size_available = 1 + while gcd % (largest_tile_size_available * 2) == 0: + largest_tile_size_available *= 2 + aspect_ratio = p.width / p.height + with split_attention(vae_model, aspect_ratio=aspect_ratio, tile_size=min(largest_tile_size_available, 128), disable=not shared.opts.hypertile_split_vae_attn): + with split_attention(unet_object, aspect_ratio=aspect_ratio, tile_size=min(largest_tile_size_available, 256), swap_size=2, disable=not shared.opts.hypertile_split_unet_attn): + flush() + samples_ddim = p.sample(conditioning=p.c, unconditional_conditioning=p.uc, seeds=p.seeds, subseeds=p.subseeds, subseed_strength=p.subseed_strength, prompts=p.prompts) if getattr(samples_ddim, 'already_decoded', False): x_samples_ddim = samples_ddim else: if opts.sd_vae_decode_method != 'Full': p.extra_generation_params['VAE Decoder'] = opts.sd_vae_decode_method - - x_samples_ddim = decode_latent_batch(p.sd_model, samples_ddim, target_device=devices.cpu, check_for_nans=True) + with split_attention(vae_model, aspect_ratio=aspect_ratio, tile_size=min(largest_tile_size_available, 128), disable=not shared.opts.hypertile_split_vae_attn): + flush() + x_samples_ddim = decode_latent_batch(p.sd_model, samples_ddim, target_device=devices.cpu, check_for_nans=True) x_samples_ddim = torch.stack(x_samples_ddim).float() x_samples_ddim = torch.clamp((x_samples_ddim + 1.0) / 2.0, min=0.0, max=1.0) @@ -980,6 +1000,7 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed: if opts.grid_save: images.save_image(grid, p.outpath_grids, "grid", p.all_seeds[0], p.all_prompts[0], opts.grid_format, info=infotext(use_main_prompt=True), short_filename=not opts.grid_extended_filename, p=p, grid=True) + random.setstate(saved_rng_state) if not p.disable_extra_networks and p.extra_network_data: extra_networks.deactivate(p, p.extra_network_data) diff --git a/modules/shared_options.py b/modules/shared_options.py index d40db5306..d96502656 100644 --- a/modules/shared_options.py +++ b/modules/shared_options.py @@ -200,6 +200,8 @@ options_templates.update(options_section(('optimizations', "Optimizations"), { "pad_cond_uncond": OptionInfo(False, "Pad prompt/negative prompt to be same length", infotext='Pad conds').info("improves performance when prompt and negative prompt have different lengths; changes seeds"), "persistent_cond_cache": OptionInfo(True, "Persistent cond cache").info("do not recalculate conds from prompts if prompts have not changed since previous calculation"), "batch_cond_uncond": OptionInfo(True, "Batch cond/uncond").info("do both conditional and unconditional denoising in one batch; uses a bit more VRAM during sampling, but improves speed; previously this was controlled by --always-batch-cond-uncond comandline argument"), + "hypertile_split_unet_attn" : OptionInfo(False, "Split attention in Unet with HyperTile").link("Github", "https://github.com/tfernd/HyperTile").info("improves performance; changes behavior, but deterministic"), + "hypertile_split_vae_attn": OptionInfo(False, "Split attention in VAE with HyperTile").link("Github", "https://github.com/tfernd/HyperTile").info("improves performance; changes behavior, but deterministic"), })) options_templates.update(options_section(('compatibility', "Compatibility"), { From 7af576e745c79a9539e40bc158e695192ae79f25 Mon Sep 17 00:00:00 2001 From: wfjsw Date: Sat, 11 Nov 2023 10:46:47 -0600 Subject: [PATCH 043/139] remove the assumption of same name --- modules/scripts.py | 81 +++++++++++++++++----------------------------- 1 file changed, 30 insertions(+), 51 deletions(-) diff --git a/modules/scripts.py b/modules/scripts.py index 7cdf288df..7ad222451 100644 --- a/modules/scripts.py +++ b/modules/scripts.py @@ -335,15 +335,9 @@ def list_scripts(scriptdirname, extension, *, include_extensions=True): for ext in extensions.active(): extension_scripts_list = ext.list_files(scriptdirname, extension) for extension_script in extension_scripts_list: - # this is built on the assumption that script name is unique. - # I think bad thing is gonna happen if name collide in the current implementation anyway, but we - # will need to refactor here if this assumption is broken later on. - if extension_script.filename in script_dependency_map: - errors.report(f"Duplicate script name \"{extension_script.filename}\" found in extensions " - f"\"{ext.name}\" and \"{script_dependency_map[extension_script.filename]['extension_dirname'] or 'builtin'}\". " - f"The current loading file will be discarded.", exc_info=False) - continue - + script_canonical_name = ext.canonical_name + "/" + extension_script.filename + if ext.is_builtin: + script_canonical_name = "builtin/" + script_canonical_name relative_path = scriptdirname + "/" + extension_script.filename requires = None @@ -359,7 +353,7 @@ def list_scripts(scriptdirname, extension, *, include_extensions=True): load_after = [x.strip() for x in load_after.split(',')] if load_after else [] load_before = [x.strip() for x in load_before.split(',')] if load_before else [] - script_dependency_map[extension_script.filename] = { + script_dependency_map[script_canonical_name] = { "extension": ext.canonical_name, "extension_dirname": ext.name, "script_file": extension_script, @@ -374,60 +368,45 @@ def list_scripts(scriptdirname, extension, *, include_extensions=True): for ext in extensions.active(): loaded_extensions.add(ext.canonical_name) - for script_filename, script_data in script_dependency_map.items(): + for script_canonical_name, script_data in script_dependency_map.items(): # load before requires inverse dependency # in this case, append the script name into the load_after list of the specified script for load_before_script in script_data['load_before']: - if load_before_script.startswith('ext:'): - # if this requires an extension to be loaded before - required_extension = load_before_script[4:] + # if this requires an individual script to be loaded before + if load_before_script in script_dependency_map: + script_dependency_map[load_before_script]['load_after'].append(script_canonical_name) + elif load_before_script in loaded_extensions: for _, script_data2 in script_dependency_map.items(): - if script_data2['extension'] == required_extension: - script_data2['load_after'].append(script_filename) + if script_data2['extension'] == load_before_script: + script_data2['load_after'].append(script_canonical_name) break - else: - # if this requires an individual script to be loaded before - if load_before_script in script_dependency_map: - script_dependency_map[load_before_script]['load_after'].append(script_filename) # resolve extension name in load_after lists - for load_after_script in script_data['load_after']: - if load_after_script.startswith('ext:'): - # if this requires an extension to be loaded after - required_extension = load_after_script[4:] - for script_file_name2, script_data2 in script_dependency_map.items(): - if script_data2['extension'] == required_extension: - script_data['load_after'].append(script_file_name2) - - # remove all extension names in load_after lists - script_data['load_after'] = [x for x in script_data['load_after'] if not x.startswith('ext:')] + for load_after_script in list(script_data['load_after']): + if load_after_script not in script_dependency_map and load_after_script in loaded_extensions: + script_data['load_after'].remove(load_after_script) + for script_canonical_name2, script_data2 in script_dependency_map.items(): + if script_data2['extension'] == load_after_script: + script_data['load_after'].remove(script_canonical_name2) + break # build the DAG sorter = TopologicalSorter() - for script_filename, script_data in script_dependency_map.items(): + for script_canonical_name, script_data in script_dependency_map.items(): requirement_met = True for required_script in script_data['requires']: - if required_script.startswith('ext:'): - # if this requires an extension to be installed - required_extension = required_script[4:] - if required_extension not in loaded_extensions: - errors.report(f"Script \"{script_filename}\" requires extension \"{required_extension}\" to " - f"be loaded, but it is not. Skipping.", - exc_info=False) - requirement_met = False - break - else: - # if this requires an individual script to be loaded - if required_script not in script_dependency_map: - errors.report(f"Script \"{script_filename}\" requires script \"{required_script}\" to " - f"be loaded, but it is not. Skipping.", - exc_info=False) - requirement_met = False - break + # if this requires an individual script to be loaded + if required_script not in script_dependency_map and required_script not in loaded_extensions: + errors.report(f"Script \"{script_canonical_name}\" " + f"requires \"{required_script}\" to " + f"be loaded, but it is not. Skipping.", + exc_info=False) + requirement_met = False + break if not requirement_met: continue - sorter.add(script_filename, *script_data['load_after']) + sorter.add(script_canonical_name, *script_data['load_after']) # sort the scripts try: @@ -436,8 +415,8 @@ def list_scripts(scriptdirname, extension, *, include_extensions=True): errors.report("Cycle detected in script dependencies. Scripts will load in ascending order.", exc_info=True) ordered_script = script_dependency_map.keys() - for script_filename in ordered_script: - script_data = script_dependency_map[script_filename] + for script_canonical_name in ordered_script: + script_data = script_dependency_map[script_canonical_name] scripts_list.append(script_data['script_file']) scripts_list = [x for x in scripts_list if os.path.splitext(x.path)[1].lower() == extension and os.path.isfile(x.path)] From 520e52f846892cc2b207b738b4180fa863c7b38f Mon Sep 17 00:00:00 2001 From: wfjsw Date: Sat, 11 Nov 2023 10:58:26 -0600 Subject: [PATCH 044/139] allow comma and whitespace as separator --- modules/extensions.py | 9 ++++++--- modules/scripts.py | 6 +++--- 2 files changed, 9 insertions(+), 6 deletions(-) diff --git a/modules/extensions.py b/modules/extensions.py index 7583a3b03..795af996e 100644 --- a/modules/extensions.py +++ b/modules/extensions.py @@ -2,6 +2,7 @@ import configparser import functools import os import threading +import re from modules import shared, errors, cache, scripts from modules.gitpython_hack import Repo @@ -48,7 +49,8 @@ class Extension: config.read(os.path.join(self.path, "sd_webui_metadata.ini")) return config except Exception: - errors.report(f"Error reading sd_webui_metadata.ini for extension {self.canonical_name}.", exc_info=True) + errors.report(f"Error reading sd_webui_metadata.ini for extension {self.canonical_name}.", + exc_info=True) return None def to_dict(self): @@ -70,6 +72,7 @@ class Extension: self.do_read_info_from_repo() return self.to_dict() + try: d = cache.cached_data_for_file('extensions-git', self.name, os.path.join(self.path, ".git"), read_from_repo) self.from_dict(d) @@ -194,8 +197,8 @@ def list_extensions(): f"The current loading extension will be discarded.", exc_info=False) continue - # we want to wash the data to lowercase and remove whitespaces just in case - requires = [x.strip() for x in requires.lower().split(',')] if requires else [] + # both "," and " " are accepted as separator + requires = list(filter(None, re.split(r"[,\s]+", requires.lower()))) if requires else [] extension_dependency_map[canonical_name] = { "dirname": extension_dirname, diff --git a/modules/scripts.py b/modules/scripts.py index 7ad222451..5dd0555dd 100644 --- a/modules/scripts.py +++ b/modules/scripts.py @@ -349,9 +349,9 @@ def list_scripts(scriptdirname, extension, *, include_extensions=True): load_before = ext.metadata.get(relative_path, "Before", fallback=None) load_after = ext.metadata.get(relative_path, "After", fallback=None) - requires = [x.strip() for x in requires.split(',')] if requires else [] - load_after = [x.strip() for x in load_after.split(',')] if load_after else [] - load_before = [x.strip() for x in load_before.split(',')] if load_before else [] + requires = list(filter(None, re.split(r"[,\s]+", requires.lower()))) if requires else [] + load_after = list(filter(None, re.split(r"[,\s]+", load_after.lower()))) if load_after else [] + load_before = list(filter(None, re.split(r"[,\s]+", load_before.lower()))) if load_before else [] script_dependency_map[script_canonical_name] = { "extension": ext.canonical_name, From 48d6102b3105bb0179c8eab14ec7930945aca326 Mon Sep 17 00:00:00 2001 From: wfjsw Date: Sat, 11 Nov 2023 11:17:26 -0600 Subject: [PATCH 045/139] fix --- modules/extensions.py | 1 - 1 file changed, 1 deletion(-) diff --git a/modules/extensions.py b/modules/extensions.py index 795af996e..5536db3ea 100644 --- a/modules/extensions.py +++ b/modules/extensions.py @@ -183,7 +183,6 @@ def list_extensions(): config.read(os.path.join(path, "sd_webui_metadata.ini")) canonical_name = config.get("Extension", "Name", fallback=canonical_name) requires = config.get("Extension", "Requires", fallback=None) - continue except Exception: errors.report(f"Error reading sd_webui_metadata.ini for extension {extension_dirname}. " f"Will load regardless.", exc_info=True) From 3bb32befe9523a6acefbab7fe099f91660f41ea9 Mon Sep 17 00:00:00 2001 From: wfjsw Date: Sat, 11 Nov 2023 11:58:19 -0600 Subject: [PATCH 046/139] bug fix --- modules/scripts.py | 25 ++++++++++++++++++------- 1 file changed, 18 insertions(+), 7 deletions(-) diff --git a/modules/scripts.py b/modules/scripts.py index 5dd0555dd..b1f4504a5 100644 --- a/modules/scripts.py +++ b/modules/scripts.py @@ -322,6 +322,9 @@ def list_scripts(scriptdirname, extension, *, include_extensions=True): root_script_basedir = os.path.join(paths.script_path, scriptdirname) if os.path.exists(root_script_basedir): for filename in sorted(os.listdir(root_script_basedir)): + if not os.path.isfile(os.path.join(root_script_basedir, filename)): + continue + script_dependency_map[filename] = { "extension": None, "extension_dirname": None, @@ -335,19 +338,27 @@ def list_scripts(scriptdirname, extension, *, include_extensions=True): for ext in extensions.active(): extension_scripts_list = ext.list_files(scriptdirname, extension) for extension_script in extension_scripts_list: + if not os.path.isfile(extension_script.path): + continue + script_canonical_name = ext.canonical_name + "/" + extension_script.filename if ext.is_builtin: script_canonical_name = "builtin/" + script_canonical_name relative_path = scriptdirname + "/" + extension_script.filename - requires = None - load_before = None - load_after = None + requires = '' + load_before = '' + load_after = '' if ext.metadata is not None: - requires = ext.metadata.get(relative_path, "Requires", fallback=None) - load_before = ext.metadata.get(relative_path, "Before", fallback=None) - load_after = ext.metadata.get(relative_path, "After", fallback=None) + requires = ext.metadata.get(relative_path, "Requires", fallback='') + load_before = ext.metadata.get(relative_path, "Before", fallback='') + load_after = ext.metadata.get(relative_path, "After", fallback='') + + # propagate directory level metadata + requires = requires + ',' + ext.metadata.get(scriptdirname, "Requires", fallback='') + load_before = load_before + ',' + ext.metadata.get(scriptdirname, "Before", fallback='') + load_after = load_after + ',' + ext.metadata.get(scriptdirname, "After", fallback='') requires = list(filter(None, re.split(r"[,\s]+", requires.lower()))) if requires else [] load_after = list(filter(None, re.split(r"[,\s]+", load_after.lower()))) if load_after else [] @@ -387,7 +398,7 @@ def list_scripts(scriptdirname, extension, *, include_extensions=True): script_data['load_after'].remove(load_after_script) for script_canonical_name2, script_data2 in script_dependency_map.items(): if script_data2['extension'] == load_after_script: - script_data['load_after'].remove(script_canonical_name2) + script_data['load_after'].append(script_canonical_name2) break # build the DAG From f6762d2ad95e3de39fc900b3fd528310e512831f Mon Sep 17 00:00:00 2001 From: Tom Haelbich <65122811+h43lb1t0@users.noreply.github.com> Date: Sun, 12 Nov 2023 14:14:16 +0100 Subject: [PATCH 047/139] dir buttons start with / so only the correct dir will be shown and not dirs with a substrings as name from the dir --- modules/ui_extra_networks.py | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/modules/ui_extra_networks.py b/modules/ui_extra_networks.py index 063bd7b80..43a94b74b 100644 --- a/modules/ui_extra_networks.py +++ b/modules/ui_extra_networks.py @@ -138,8 +138,9 @@ class ExtraNetworksPage: continue subdir = os.path.abspath(x)[len(parentdir):].replace("\\", "/") - while subdir.startswith("/"): - subdir = subdir[1:] + + if not subdir.startswith("/"): + subdir = "/" + subdir is_empty = len(os.listdir(x)) == 0 if not is_empty and not subdir.endswith("/"): From 8048f36072c8a281b8c8c79235df63a748ab7361 Mon Sep 17 00:00:00 2001 From: missionfloyd Date: Sun, 12 Nov 2023 17:12:50 -0700 Subject: [PATCH 048/139] Lint --- modules/ui_extra_networks.py | 1 - 1 file changed, 1 deletion(-) diff --git a/modules/ui_extra_networks.py b/modules/ui_extra_networks.py index 43a94b74b..bd6732856 100644 --- a/modules/ui_extra_networks.py +++ b/modules/ui_extra_networks.py @@ -138,7 +138,6 @@ class ExtraNetworksPage: continue subdir = os.path.abspath(x)[len(parentdir):].replace("\\", "/") - if not subdir.startswith("/"): subdir = "/" + subdir From 94e966956666ba13b368aaf781628085e3d4f7e3 Mon Sep 17 00:00:00 2001 From: kaalibro Date: Mon, 13 Nov 2023 14:51:06 +0600 Subject: [PATCH 049/139] Fixes generation restart not working for some users when 'Ctrl+Enter' is pressed --- script.js | 15 ++++++++++++--- 1 file changed, 12 insertions(+), 3 deletions(-) diff --git a/script.js b/script.js index 5f6ee3542..c0e678ea7 100644 --- a/script.js +++ b/script.js @@ -133,9 +133,18 @@ document.addEventListener('keydown', function(e) { if (isEnter && isModifierKey) { if (interruptButton.style.display === 'block') { interruptButton.click(); - setTimeout(function() { - generateButton.click(); - }, 500); + const callback = (mutationList) => { + for (const mutation of mutationList) { + if (mutation.type === 'attributes' && mutation.attributeName === 'style') { + if (interruptButton.style.display === 'none') { + generateButton.click(); + observer.disconnect(); + } + } + } + }; + const observer = new MutationObserver(callback); + observer.observe(interruptButton, {attributes: true}); } else { generateButton.click(); } From c1c816006e47f3b7dcf1512594fd31817242e7fa Mon Sep 17 00:00:00 2001 From: kaalibro Date: Mon, 13 Nov 2023 22:01:52 +0600 Subject: [PATCH 050/139] Adds 'Path' sorting for Extra network cards --- modules/shared_options.py | 2 +- modules/ui_extra_networks.py | 3 ++- 2 files changed, 3 insertions(+), 2 deletions(-) diff --git a/modules/shared_options.py b/modules/shared_options.py index d40db5306..8fc7ef1d2 100644 --- a/modules/shared_options.py +++ b/modules/shared_options.py @@ -235,7 +235,7 @@ options_templates.update(options_section(('extra_networks', "Extra Networks"), { "extra_networks_card_height": OptionInfo(0, "Card height for Extra Networks").info("in pixels"), "extra_networks_card_text_scale": OptionInfo(1.0, "Card text scale", gr.Slider, {"minimum": 0.0, "maximum": 2.0, "step": 0.01}).info("1 = original size"), "extra_networks_card_show_desc": OptionInfo(True, "Show description on card"), - "extra_networks_card_order_field": OptionInfo("Name", "Default order field for Extra Networks cards", gr.Dropdown, {"choices": ['Name', 'Date Created', 'Date Modified']}).needs_reload_ui(), + "extra_networks_card_order_field": OptionInfo("Path", "Default order field for Extra Networks cards", gr.Dropdown, {"choices": ['Path', 'Name', 'Date Created', 'Date Modified']}).needs_reload_ui(), "extra_networks_card_order": OptionInfo("Ascending", "Default order for Extra Networks cards", gr.Dropdown, {"choices": ['Ascending', 'Descending']}).needs_reload_ui(), "extra_networks_add_text_separator": OptionInfo(" ", "Extra networks separator").info("extra text to add before <...> when adding extra network to prompt"), "ui_extra_networks_tab_reorder": OptionInfo("", "Extra networks tab order").needs_reload_ui(), diff --git a/modules/ui_extra_networks.py b/modules/ui_extra_networks.py index 7907cd63f..f03e20337 100644 --- a/modules/ui_extra_networks.py +++ b/modules/ui_extra_networks.py @@ -279,6 +279,7 @@ class ExtraNetworksPage: "date_created": int(stat.st_ctime or 0), "date_modified": int(stat.st_mtime or 0), "name": pth.name.lower(), + "path": str(pth.parent).lower(), } def find_preview(self, path): @@ -382,7 +383,7 @@ def create_ui(interface: gr.Blocks, unrelated_tabs, tabname): related_tabs.append(tab) edit_search = gr.Textbox('', show_label=False, elem_id=tabname+"_extra_search", elem_classes="search", placeholder="Search...", visible=False, interactive=True) - dropdown_sort = gr.Dropdown(choices=['Name', 'Date Created', 'Date Modified', ], value=shared.opts.extra_networks_card_order_field, elem_id=tabname+"_extra_sort", elem_classes="sort", multiselect=False, visible=False, show_label=False, interactive=True, label=tabname+"_extra_sort_order") + dropdown_sort = gr.Dropdown(choices=['Path', 'Name', 'Date Created', 'Date Modified', ], value=shared.opts.extra_networks_card_order_field, elem_id=tabname+"_extra_sort", elem_classes="sort", multiselect=False, visible=False, show_label=False, interactive=True, label=tabname+"_extra_sort_order") button_sortorder = ToolButton(switch_values_symbol, elem_id=tabname+"_extra_sortorder", elem_classes=["sortorder"] + ([] if shared.opts.extra_networks_card_order == "Ascending" else ["sortReverse"]), visible=False, tooltip="Invert sort order") button_refresh = gr.Button('Refresh', elem_id=tabname+"_extra_refresh", visible=False) checkbox_show_dirs = gr.Checkbox(True, label='Show dirs', elem_id=tabname+"_extra_show_dirs", elem_classes="show-dirs", visible=False) From a292d2c47f51fc71cc186709bdf3706f0944b7d6 Mon Sep 17 00:00:00 2001 From: AngelBottomless Date: Wed, 15 Nov 2023 14:26:37 +0900 Subject: [PATCH 051/139] hotfix: call shared.state.end() after postprocessing done --- modules/postprocessing.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/modules/postprocessing.py b/modules/postprocessing.py index cf04d38b0..fd0c0cc99 100644 --- a/modules/postprocessing.py +++ b/modules/postprocessing.py @@ -78,7 +78,7 @@ def run_postprocessing(extras_mode, image, image_folder, input_dir, output_dir, image_data.close() devices.torch_gc() - + shared.state.end() return outputs, ui_common.plaintext_to_html(infotext), '' From b29fc6d4de8812b25c520a46676cda13c3fe64ca Mon Sep 17 00:00:00 2001 From: aria1th <35677394+aria1th@users.noreply.github.com> Date: Sat, 11 Nov 2023 23:43:13 +0900 Subject: [PATCH 052/139] Implement Hypertile Co-Authored-By: Kieran Hunt --- modules/hypertile.py | 333 ++++++++++++++++++++++++++++++++++++++++++ modules/processing.py | 65 ++++----- 2 files changed, 358 insertions(+), 40 deletions(-) create mode 100644 modules/hypertile.py diff --git a/modules/hypertile.py b/modules/hypertile.py new file mode 100644 index 000000000..ab1c74c02 --- /dev/null +++ b/modules/hypertile.py @@ -0,0 +1,333 @@ +""" +Hypertile module for splitting attention layers in SD-1.5 U-Net and SD-1.5 VAE +Warn : The patch works well only if the input image has a width and height that are multiples of 128 +Author : @tfernd Github : https://github.com/tfernd/HyperTile +""" + +from __future__ import annotations +from typing import Callable +from typing_extensions import Literal + +import logging +from functools import wraps, cache +from contextlib import contextmanager + +import math +import torch.nn as nn +import random + +from einops import rearrange + +# TODO add SD-XL layers +DEPTH_LAYERS = { + 0: [ + # SD 1.5 U-Net (diffusers) + "down_blocks.0.attentions.0.transformer_blocks.0.attn1", + "down_blocks.0.attentions.1.transformer_blocks.0.attn1", + "up_blocks.3.attentions.0.transformer_blocks.0.attn1", + "up_blocks.3.attentions.1.transformer_blocks.0.attn1", + "up_blocks.3.attentions.2.transformer_blocks.0.attn1", + # SD 1.5 U-Net (ldm) + "input_blocks.1.1.transformer_blocks.0.attn1", + "input_blocks.2.1.transformer_blocks.0.attn1", + "output_blocks.9.1.transformer_blocks.0.attn1", + "output_blocks.10.1.transformer_blocks.0.attn1", + "output_blocks.11.1.transformer_blocks.0.attn1", + # SD 1.5 VAE + "decoder.mid_block.attentions.0", + ], + 1: [ + # SD 1.5 U-Net (diffusers) + "down_blocks.1.attentions.0.transformer_blocks.0.attn1", + "down_blocks.1.attentions.1.transformer_blocks.0.attn1", + "up_blocks.2.attentions.0.transformer_blocks.0.attn1", + "up_blocks.2.attentions.1.transformer_blocks.0.attn1", + "up_blocks.2.attentions.2.transformer_blocks.0.attn1", + # SD 1.5 U-Net (ldm) + "input_blocks.4.1.transformer_blocks.0.attn1", + "input_blocks.5.1.transformer_blocks.0.attn1", + "output_blocks.6.1.transformer_blocks.0.attn1", + "output_blocks.7.1.transformer_blocks.0.attn1", + "output_blocks.8.1.transformer_blocks.0.attn1", + ], + 2: [ + # SD 1.5 U-Net (diffusers) + "down_blocks.2.attentions.0.transformer_blocks.0.attn1", + "down_blocks.2.attentions.1.transformer_blocks.0.attn1", + "up_blocks.1.attentions.0.transformer_blocks.0.attn1", + "up_blocks.1.attentions.1.transformer_blocks.0.attn1", + "up_blocks.1.attentions.2.transformer_blocks.0.attn1", + # SD 1.5 U-Net (ldm) + "input_blocks.7.1.transformer_blocks.0.attn1", + "input_blocks.8.1.transformer_blocks.0.attn1", + "output_blocks.3.1.transformer_blocks.0.attn1", + "output_blocks.4.1.transformer_blocks.0.attn1", + "output_blocks.5.1.transformer_blocks.0.attn1", + ], + 3: [ + # SD 1.5 U-Net (diffusers) + "mid_block.attentions.0.transformer_blocks.0.attn1", + # SD 1.5 U-Net (ldm) + "middle_block.1.transformer_blocks.0.attn1", + ], +} +# XL layers, thanks for GitHub@gel-crabs for the help +DEPTH_LAYERS_XL = { + 0: [ + # SD 1.5 U-Net (diffusers) + "down_blocks.0.attentions.0.transformer_blocks.0.attn1", + "down_blocks.0.attentions.1.transformer_blocks.0.attn1", + "up_blocks.3.attentions.0.transformer_blocks.0.attn1", + "up_blocks.3.attentions.1.transformer_blocks.0.attn1", + "up_blocks.3.attentions.2.transformer_blocks.0.attn1", + # SD 1.5 U-Net (ldm) + "input_blocks.4.1.transformer_blocks.0.attn1", + "input_blocks.5.1.transformer_blocks.0.attn1", + "output_blocks.3.1.transformer_blocks.0.attn1", + "output_blocks.4.1.transformer_blocks.0.attn1", + "output_blocks.5.1.transformer_blocks.0.attn1", + # SD 1.5 VAE + "decoder.mid_block.attentions.0", + "decoder.mid.attn_1", + ], + 1: [ + # SD 1.5 U-Net (diffusers) + #"down_blocks.1.attentions.0.transformer_blocks.0.attn1", + #"down_blocks.1.attentions.1.transformer_blocks.0.attn1", + #"up_blocks.2.attentions.0.transformer_blocks.0.attn1", + #"up_blocks.2.attentions.1.transformer_blocks.0.attn1", + #"up_blocks.2.attentions.2.transformer_blocks.0.attn1", + # SD 1.5 U-Net (ldm) + "input_blocks.4.1.transformer_blocks.1.attn1", + "input_blocks.5.1.transformer_blocks.1.attn1", + "output_blocks.3.1.transformer_blocks.1.attn1", + "output_blocks.4.1.transformer_blocks.1.attn1", + "output_blocks.5.1.transformer_blocks.1.attn1", + "input_blocks.7.1.transformer_blocks.0.attn1", + "input_blocks.8.1.transformer_blocks.0.attn1", + "output_blocks.0.1.transformer_blocks.0.attn1", + "output_blocks.1.1.transformer_blocks.0.attn1", + "output_blocks.2.1.transformer_blocks.0.attn1", + "input_blocks.7.1.transformer_blocks.1.attn1", + "input_blocks.8.1.transformer_blocks.1.attn1", + "output_blocks.0.1.transformer_blocks.1.attn1", + "output_blocks.1.1.transformer_blocks.1.attn1", + "output_blocks.2.1.transformer_blocks.1.attn1", + "input_blocks.7.1.transformer_blocks.2.attn1", + "input_blocks.8.1.transformer_blocks.2.attn1", + "output_blocks.0.1.transformer_blocks.2.attn1", + "output_blocks.1.1.transformer_blocks.2.attn1", + "output_blocks.2.1.transformer_blocks.2.attn1", + "input_blocks.7.1.transformer_blocks.3.attn1", + "input_blocks.8.1.transformer_blocks.3.attn1", + "output_blocks.0.1.transformer_blocks.3.attn1", + "output_blocks.1.1.transformer_blocks.3.attn1", + "output_blocks.2.1.transformer_blocks.3.attn1", + "input_blocks.7.1.transformer_blocks.4.attn1", + "input_blocks.8.1.transformer_blocks.4.attn1", + "output_blocks.0.1.transformer_blocks.4.attn1", + "output_blocks.1.1.transformer_blocks.4.attn1", + "output_blocks.2.1.transformer_blocks.4.attn1", + "input_blocks.7.1.transformer_blocks.5.attn1", + "input_blocks.8.1.transformer_blocks.5.attn1", + "output_blocks.0.1.transformer_blocks.5.attn1", + "output_blocks.1.1.transformer_blocks.5.attn1", + "output_blocks.2.1.transformer_blocks.5.attn1", + "input_blocks.7.1.transformer_blocks.6.attn1", + "input_blocks.8.1.transformer_blocks.6.attn1", + "output_blocks.0.1.transformer_blocks.6.attn1", + "output_blocks.1.1.transformer_blocks.6.attn1", + "output_blocks.2.1.transformer_blocks.6.attn1", + "input_blocks.7.1.transformer_blocks.7.attn1", + "input_blocks.8.1.transformer_blocks.7.attn1", + "output_blocks.0.1.transformer_blocks.7.attn1", + "output_blocks.1.1.transformer_blocks.7.attn1", + "output_blocks.2.1.transformer_blocks.7.attn1", + "input_blocks.7.1.transformer_blocks.8.attn1", + "input_blocks.8.1.transformer_blocks.8.attn1", + "output_blocks.0.1.transformer_blocks.8.attn1", + "output_blocks.1.1.transformer_blocks.8.attn1", + "output_blocks.2.1.transformer_blocks.8.attn1", + "input_blocks.7.1.transformer_blocks.9.attn1", + "input_blocks.8.1.transformer_blocks.9.attn1", + "output_blocks.0.1.transformer_blocks.9.attn1", + "output_blocks.1.1.transformer_blocks.9.attn1", + "output_blocks.2.1.transformer_blocks.9.attn1", + ], + 2: [ + # SD 1.5 U-Net (diffusers) + "mid_block.attentions.0.transformer_blocks.0.attn1", + # SD 1.5 U-Net (ldm) + "middle_block.1.transformer_blocks.0.attn1", + "middle_block.1.transformer_blocks.1.attn1", + "middle_block.1.transformer_blocks.2.attn1", + "middle_block.1.transformer_blocks.3.attn1", + "middle_block.1.transformer_blocks.4.attn1", + "middle_block.1.transformer_blocks.5.attn1", + "middle_block.1.transformer_blocks.6.attn1", + "middle_block.1.transformer_blocks.7.attn1", + "middle_block.1.transformer_blocks.8.attn1", + "middle_block.1.transformer_blocks.9.attn1", + ], +} + + +RNG_INSTANCE = random.Random() + +def random_divisor(value: int, min_value: int, /, max_options: int = 1) -> int: + """ + Returns a random divisor of value that + x * min_value <= value + if max_options is 1, the behavior is deterministic + """ + min_value = min(min_value, value) + + # All big divisors of value (inclusive) + divisors = [i for i in range(min_value, value + 1) if value % i == 0] # divisors in small -> big order + + ns = [value // i for i in divisors[:max_options]] # has at least 1 element # big -> small order + + idx = RNG_INSTANCE.randint(0, len(ns) - 1) + + return ns[idx] + +def set_hypertile_seed(seed: int) -> None: + RNG_INSTANCE.seed(seed) + +def largest_tile_size_available(width:int, height:int) -> int: + """ + Calculates the largest tile size available for a given width and height + Tile size is always a power of 2 + """ + gcd = math.gcd(width, height) + largest_tile_size_available = 1 + while gcd % (largest_tile_size_available * 2) == 0: + largest_tile_size_available *= 2 + return largest_tile_size_available + +def iterative_closest_divisors(hw:int, aspect_ratio:float) -> tuple[int, int]: + """ + Finds h and w such that h*w = hw and h/w = aspect_ratio + We check all possible divisors of hw and return the closest to the aspect ratio + """ + divisors = [i for i in range(2, hw + 1) if hw % i == 0] # all divisors of hw + pairs = [(i, hw // i) for i in divisors] # all pairs of divisors of hw + ratios = [w/h for h, w in pairs] # all ratios of pairs of divisors of hw + closest_ratio = min(ratios, key=lambda x: abs(x - aspect_ratio)) # closest ratio to aspect_ratio + closest_pair = pairs[ratios.index(closest_ratio)] # closest pair of divisors to aspect_ratio + return closest_pair + +@cache +def find_hw_candidates(hw:int, aspect_ratio:float) -> tuple[int, int]: + """ + Finds h and w such that h*w = hw and h/w = aspect_ratio + """ + h, w = round(math.sqrt(hw * aspect_ratio)), round(math.sqrt(hw / aspect_ratio)) + # find h and w such that h*w = hw and h/w = aspect_ratio + if h * w != hw: + w_candidate = hw / h + # check if w is an integer + if not w_candidate.is_integer(): + h_candidate = hw / w + # check if h is an integer + if not h_candidate.is_integer(): + return iterative_closest_divisors(hw, aspect_ratio) + else: + h = int(h_candidate) + else: + w = int(w_candidate) + return h, w + +@contextmanager +def split_attention( + layer: nn.Module, + /, + aspect_ratio: float, # width/height + tile_size: int = 128, # 128 for VAE + swap_size: int = 1, # 1 for VAE + *, + disable: bool = False, + max_depth: Literal[0, 1, 2, 3] = 0, # ! Try 0 or 1 + scale_depth: bool = True, # scale the tile-size depending on the depth + is_sdxl: bool = False, # is the model SD-XL +): + # Hijacks AttnBlock from ldm and Attention from diffusers + + if disable: + logging.info(f"Attention for {layer.__class__.__qualname__} not splitted") + yield + return + + latent_tile_size = max(128, tile_size) // 8 + + def self_attn_forward(forward: Callable, depth: int, layer_name: str, module: nn.Module) -> Callable: + @wraps(forward) + def wrapper(*args, **kwargs): + x = args[0] + + # VAE + if x.ndim == 4: + b, c, h, w = x.shape + + nh = random_divisor(h, latent_tile_size, swap_size) + nw = random_divisor(w, latent_tile_size, swap_size) + + if nh * nw > 1: + x = rearrange(x, "b c (nh h) (nw w) -> (b nh nw) c h w", nh=nh, nw=nw) # split into nh * nw tiles + + out = forward(x, *args[1:], **kwargs) + + if nh * nw > 1: + out = rearrange(out, "(b nh nw) c h w -> b c (nh h) (nw w)", nh=nh, nw=nw) + + # U-Net + else: + hw: int = x.size(1) + h, w = find_hw_candidates(hw, aspect_ratio) + assert h * w == hw, f"Invalid aspect ratio {aspect_ratio} for input of shape {x.shape}, hw={hw}, h={h}, w={w}" + + factor = 2**depth if scale_depth else 1 + nh = random_divisor(h, latent_tile_size * factor, swap_size) + nw = random_divisor(w, latent_tile_size * factor, swap_size) + + module._split_sizes_hypertile.append((nh, nw)) # type: ignore + + if nh * nw > 1: + x = rearrange(x, "b (nh h nw w) c -> (b nh nw) (h w) c", h=h // nh, w=w // nw, nh=nh, nw=nw) + + out = forward(x, *args[1:], **kwargs) + + if nh * nw > 1: + out = rearrange(out, "(b nh nw) hw c -> b nh nw hw c", nh=nh, nw=nw) + out = rearrange(out, "b nh nw (h w) c -> b (nh h nw w) c", h=h // nh, w=w // nw) + + return out + + return wrapper + + # Handle hijacking the forward method and recovering afterwards + try: + if is_sdxl: + layers = DEPTH_LAYERS_XL + else: + layers = DEPTH_LAYERS + for depth in range(max_depth + 1): + for layer_name, module in layer.named_modules(): + if any(layer_name.endswith(try_name) for try_name in layers[depth]): + # print input shape for debugging + logging.debug(f"HyperTile hijacking attention layer at depth {depth}: {layer_name}") + # hijack + module._original_forward_hypertile = module.forward + module.forward = self_attn_forward(module.forward, depth, layer_name, module) + module._split_sizes_hypertile = [] + yield + finally: + for layer_name, module in layer.named_modules(): + # remove hijack + if hasattr(module, "_original_forward_hypertile"): + if module._split_sizes_hypertile: + logging.debug(f"layer {layer_name} splitted with ({module._split_sizes_hypertile})") + # recover + module.forward = module._original_forward_hypertile + del module._original_forward_hypertile + del module._split_sizes_hypertile diff --git a/modules/processing.py b/modules/processing.py index e23095343..e19a09a3c 100644 --- a/modules/processing.py +++ b/modules/processing.py @@ -24,6 +24,7 @@ from modules.shared import opts, cmd_opts, state import modules.shared as shared import modules.paths as paths import modules.face_restoration +from modules.hypertile import split_attention, set_hypertile_seed, largest_tile_size_available import modules.images as images import modules.styles import modules.sd_models as sd_models @@ -799,17 +800,6 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed: infotexts = [] output_images = [] - unet_object = p.sd_model.model - vae_model = p.sd_model.first_stage_model - try: - from hyper_tile import split_attention, flush - except (ImportError, ModuleNotFoundError): # pip install git+https://github.com/tfernd/HyperTile@2ef64b2800d007d305755c33550537410310d7df - split_attention = lambda *args, **kwargs: lambda x: x # return a no-op context manager - flush = lambda: None - import random - saved_rng_state = random.getstate() - random.seed(p.seed) # hyper_tile uses random, so we need to seed it - with torch.no_grad(), p.sd_model.ema_scope(): with devices.autocast(): p.init(p.all_prompts, p.all_seeds, p.all_subseeds) @@ -871,29 +861,20 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed: p.comment(comment) p.extra_generation_params.update(model_hijack.extra_generation_params) - + set_hypertile_seed(p.seed) + # add batch size + hypertile status to information to reproduce the run if p.n_iter > 1: shared.state.job = f"Batch {n+1} out of {p.n_iter}" with devices.without_autocast() if devices.unet_needs_upcast else devices.autocast(): - # get largest tile size available, which is 2^x which is factor of gcd of p.width and p.height - gcd = math.gcd(p.width, p.height) - largest_tile_size_available = 1 - while gcd % (largest_tile_size_available * 2) == 0: - largest_tile_size_available *= 2 - aspect_ratio = p.width / p.height - with split_attention(vae_model, aspect_ratio=aspect_ratio, tile_size=min(largest_tile_size_available, 128), disable=not shared.opts.hypertile_split_vae_attn): - with split_attention(unet_object, aspect_ratio=aspect_ratio, tile_size=min(largest_tile_size_available, 256), swap_size=2, disable=not shared.opts.hypertile_split_unet_attn): - flush() - samples_ddim = p.sample(conditioning=p.c, unconditional_conditioning=p.uc, seeds=p.seeds, subseeds=p.subseeds, subseed_strength=p.subseed_strength, prompts=p.prompts) + samples_ddim = p.sample(conditioning=p.c, unconditional_conditioning=p.uc, seeds=p.seeds, subseeds=p.subseeds, subseed_strength=p.subseed_strength, prompts=p.prompts) if getattr(samples_ddim, 'already_decoded', False): x_samples_ddim = samples_ddim else: if opts.sd_vae_decode_method != 'Full': p.extra_generation_params['VAE Decoder'] = opts.sd_vae_decode_method - with split_attention(vae_model, aspect_ratio=aspect_ratio, tile_size=min(largest_tile_size_available, 128), disable=not shared.opts.hypertile_split_vae_attn): - flush() + with split_attention(p.sd_model.first_stage_model, aspect_ratio = p.width / p.height, tile_size=min(largest_tile_size_available(p.width, p.height), 128), disable=not shared.opts.hypertile_split_vae_attn, is_sdxl=shared.sd_model.is_sdxl): x_samples_ddim = decode_latent_batch(p.sd_model, samples_ddim, target_device=devices.cpu, check_for_nans=True) x_samples_ddim = torch.stack(x_samples_ddim).float() @@ -1000,7 +981,6 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed: if opts.grid_save: images.save_image(grid, p.outpath_grids, "grid", p.all_seeds[0], p.all_prompts[0], opts.grid_format, info=infotext(use_main_prompt=True), short_filename=not opts.grid_extended_filename, p=p, grid=True) - random.setstate(saved_rng_state) if not p.disable_extra_networks and p.extra_network_data: extra_networks.deactivate(p, p.extra_network_data) @@ -1161,24 +1141,25 @@ class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing): def sample(self, conditioning, unconditional_conditioning, seeds, subseeds, subseed_strength, prompts): self.sampler = sd_samplers.create_sampler(self.sampler_name, self.sd_model) - + aspect_ratio = self.width / self.height x = self.rng.next() - samples = self.sampler.sample(self, x, conditioning, unconditional_conditioning, image_conditioning=self.txt2img_image_conditioning(x)) + tile_size = largest_tile_size_available(self.width, self.height) + with split_attention(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=min(tile_size, 128), swap_size=1, disable=not shared.opts.hypertile_split_vae_attn, is_sdxl=shared.sd_model.is_sdxl): + with split_attention(self.sd_model.model, aspect_ratio=aspect_ratio, tile_size=min(tile_size, 256), swap_size=2, disable=not shared.opts.hypertile_split_unet_attn, is_sdxl=shared.sd_model.is_sdxl): + devices.torch_gc() + samples = self.sampler.sample(self, x, conditioning, unconditional_conditioning, image_conditioning=self.txt2img_image_conditioning(x)) del x - if not self.enable_hr: return samples if self.latent_scale_mode is None: - decoded_samples = torch.stack(decode_latent_batch(self.sd_model, samples, target_device=devices.cpu, check_for_nans=True)).to(dtype=torch.float32) + with split_attention(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=min(tile_size, 256), swap_size=1, disable=not shared.opts.hypertile_split_vae_attn, is_sdxl=shared.sd_model.is_sdxl): + decoded_samples = torch.stack(decode_latent_batch(self.sd_model, samples, target_device=devices.cpu, check_for_nans=True)).to(dtype=torch.float32) else: decoded_samples = None with sd_models.SkipWritingToConfig(): sd_models.reload_model_weights(info=self.hr_checkpoint_info) - - devices.torch_gc() - return self.sample_hr_pass(samples, decoded_samples, seeds, subseeds, subseed_strength, prompts) def sample_hr_pass(self, samples, decoded_samples, seeds, subseeds, subseed_strength, prompts): @@ -1186,7 +1167,6 @@ class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing): return samples self.is_hr_pass = True - target_width = self.hr_upscale_to_x target_height = self.hr_upscale_to_y @@ -1264,18 +1244,19 @@ class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing): if self.scripts is not None: self.scripts.before_hr(self) - - samples = self.sampler.sample_img2img(self, samples, noise, self.hr_c, self.hr_uc, steps=self.hr_second_pass_steps or self.steps, image_conditioning=image_conditioning) + tile_size = largest_tile_size_available(target_width, target_height) + with split_attention(self.sd_model.first_stage_model, aspect_ratio=target_width / target_height, tile_size=min(tile_size, 256), swap_size=1, disable=not opts.hypertile_split_vae_attn, is_sdxl=shared.sd_model.is_sdxl): + with split_attention(self.sd_model.model, aspect_ratio=target_width / target_height, tile_size=min(tile_size, 256), swap_size=3, max_depth=1,scale_depth=True, disable=not opts.hypertile_split_unet_attn, is_sdxl=shared.sd_model.is_sdxl): + samples = self.sampler.sample_img2img(self, samples, noise, self.hr_c, self.hr_uc, steps=self.hr_second_pass_steps or self.steps, image_conditioning=image_conditioning) sd_models.apply_token_merging(self.sd_model, self.get_token_merging_ratio()) self.sampler = None devices.torch_gc() - - decoded_samples = decode_latent_batch(self.sd_model, samples, target_device=devices.cpu, check_for_nans=True) + with split_attention(self.sd_model.first_stage_model, aspect_ratio=target_width / target_height, tile_size=min(tile_size, 256), swap_size=1, disable=not opts.hypertile_split_vae_attn, is_sdxl=shared.sd_model.is_sdxl): + decoded_samples = decode_latent_batch(self.sd_model, samples, target_device=devices.cpu, check_for_nans=True) self.is_hr_pass = False - return decoded_samples def close(self): @@ -1550,8 +1531,12 @@ class StableDiffusionProcessingImg2Img(StableDiffusionProcessing): if self.initial_noise_multiplier != 1.0: self.extra_generation_params["Noise multiplier"] = self.initial_noise_multiplier x *= self.initial_noise_multiplier - - samples = self.sampler.sample_img2img(self, self.init_latent, x, conditioning, unconditional_conditioning, image_conditioning=self.image_conditioning) + aspect_ratio = self.width / self.height + tile_size = largest_tile_size_available(self.width, self.height) + with split_attention(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=min(tile_size, 128), swap_size=1, disable=not shared.opts.hypertile_split_vae_attn, is_sdxl=shared.sd_model.is_sdxl): + with split_attention(self.sd_model.model, aspect_ratio=aspect_ratio, tile_size=min(tile_size, 256), swap_size=2, disable=not shared.opts.hypertile_split_unet_attn, is_sdxl=shared.sd_model.is_sdxl): + devices.torch_gc() + samples = self.sampler.sample_img2img(self, self.init_latent, x, conditioning, unconditional_conditioning, image_conditioning=self.image_conditioning) if self.mask is not None: samples = samples * self.nmask + self.init_latent * self.mask From af45872fdb8a66ffd6a405d99120e0bacbb4a170 Mon Sep 17 00:00:00 2001 From: aria1th <35677394+aria1th@users.noreply.github.com> Date: Wed, 15 Nov 2023 15:15:14 +0900 Subject: [PATCH 053/139] copy LDM VAE key from XL --- modules/hypertile.py | 1 + 1 file changed, 1 insertion(+) diff --git a/modules/hypertile.py b/modules/hypertile.py index ab1c74c02..32d8604cc 100644 --- a/modules/hypertile.py +++ b/modules/hypertile.py @@ -35,6 +35,7 @@ DEPTH_LAYERS = { "output_blocks.11.1.transformer_blocks.0.attn1", # SD 1.5 VAE "decoder.mid_block.attentions.0", + "decoder.mid.attn_1", ], 1: [ # SD 1.5 U-Net (diffusers) From d6d0b22e6657fc84039e82ee735a57101bfe7c17 Mon Sep 17 00:00:00 2001 From: v0xie <28695009+v0xie@users.noreply.github.com> Date: Wed, 15 Nov 2023 03:08:50 -0800 Subject: [PATCH 054/139] fix: ignore calc_scale() for COFT which has very small alpha --- extensions-builtin/Lora/network_oft.py | 16 +++++----------- 1 file changed, 5 insertions(+), 11 deletions(-) diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py index 93402bb28..c45a8d23a 100644 --- a/extensions-builtin/Lora/network_oft.py +++ b/extensions-builtin/Lora/network_oft.py @@ -99,12 +99,9 @@ class NetworkModuleOFT(network.NetworkModule): is_other_linear = type(self.sd_module) in [torch.nn.MultiheadAttention] if not is_other_linear: - #if is_other_linear and orig_weight.shape[0] != orig_weight.shape[1]: - # orig_weight=orig_weight.permute(1, 0) - oft_blocks = self.oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) - # without this line the results are significantly worse / less accurate + # ensure skew-symmetric matrix oft_blocks = oft_blocks - oft_blocks.transpose(1, 2) R = oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) @@ -118,9 +115,6 @@ class NetworkModuleOFT(network.NetworkModule): ) merged_weight = rearrange(merged_weight, 'k m ... -> (k m) ...') - #if is_other_linear and orig_weight.shape[0] != orig_weight.shape[1]: - # orig_weight=orig_weight.permute(1, 0) - updown = merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight output_shape = orig_weight.shape else: @@ -132,10 +126,10 @@ class NetworkModuleOFT(network.NetworkModule): return self.finalize_updown(updown, orig_weight, output_shape) def calc_updown(self, orig_weight): - multiplier = self.multiplier() * self.calc_scale() - #if self.is_kohya: - # return self.calc_updown_kohya(orig_weight, multiplier) - #else: + # if alpha is a very small number as in coft, calc_scale will return a almost zero number so we ignore it + #multiplier = self.multiplier() * self.calc_scale() + multiplier = self.multiplier() + return self.calc_updown_kb(orig_weight, multiplier) # override to remove the multiplier/scale factor; it's already multiplied in get_weight From eb667e715ad3eea981f6263c143ab0422e5340c9 Mon Sep 17 00:00:00 2001 From: v0xie <28695009+v0xie@users.noreply.github.com> Date: Wed, 15 Nov 2023 18:28:48 -0800 Subject: [PATCH 055/139] feat: LyCORIS/kohya OFT network support --- extensions-builtin/Lora/network_oft.py | 108 ++++++------------------- 1 file changed, 26 insertions(+), 82 deletions(-) diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py index c45a8d23a..05c378118 100644 --- a/extensions-builtin/Lora/network_oft.py +++ b/extensions-builtin/Lora/network_oft.py @@ -11,8 +11,8 @@ class ModuleTypeOFT(network.ModuleType): return None -# adapted from kohya-ss' implementation https://github.com/kohya-ss/sd-scripts/blob/main/networks/oft.py -# and KohakuBlueleaf's implementation https://github.com/KohakuBlueleaf/LyCORIS/blob/dev/lycoris/modules/diag_oft.py +# Supports both kohya-ss' implementation of COFT https://github.com/kohya-ss/sd-scripts/blob/main/networks/oft.py +# and KohakuBlueleaf's implementation of OFT/COFT https://github.com/KohakuBlueleaf/LyCORIS/blob/dev/lycoris/modules/diag_oft.py class NetworkModuleOFT(network.NetworkModule): def __init__(self, net: network.Network, weights: network.NetworkWeights): @@ -25,117 +25,61 @@ class NetworkModuleOFT(network.NetworkModule): if "oft_blocks" in weights.w.keys(): self.is_kohya = True self.oft_blocks = weights.w["oft_blocks"] # (num_blocks, block_size, block_size) - self.alpha = weights.w["alpha"] + self.alpha = weights.w["alpha"] # alpha is constraint self.dim = self.oft_blocks.shape[0] # lora dim - #self.oft_blocks = rearrange(self.oft_blocks, 'k m ... -> (k m) ...') + # LyCORIS elif "oft_diag" in weights.w.keys(): self.is_kohya = False - self.oft_blocks = weights.w["oft_diag"] # (num_blocks, block_size, block_size) - - # alpha is rank if alpha is 0 or None - if self.alpha is None: - pass - self.dim = self.oft_blocks.shape[1] # FIXME: almost certainly incorrect, assumes tensor is shape [*, m, n] - else: - raise ValueError("oft_blocks or oft_diag must be in weights dict") + self.oft_blocks = weights.w["oft_diag"] + # self.alpha is unused + self.dim = self.oft_blocks.shape[1] # (num_blocks, block_size, block_size) is_linear = type(self.sd_module) in [torch.nn.Linear, torch.nn.modules.linear.NonDynamicallyQuantizableLinear] is_conv = type(self.sd_module) in [torch.nn.Conv2d] - is_other_linear = type(self.sd_module) in [torch.nn.MultiheadAttention] + is_other_linear = type(self.sd_module) in [torch.nn.MultiheadAttention] # unsupported if is_linear: self.out_dim = self.sd_module.out_features - elif is_other_linear: - self.out_dim = self.sd_module.embed_dim elif is_conv: self.out_dim = self.sd_module.out_channels - else: - raise ValueError("sd_module must be Linear or Conv") + elif is_other_linear: + self.out_dim = self.sd_module.embed_dim if self.is_kohya: self.constraint = self.alpha * self.out_dim - self.num_blocks, self.block_size = factorization(self.out_dim, self.dim) + self.num_blocks = self.dim + self.block_size = self.out_dim // self.dim else: self.constraint = None self.block_size, self.num_blocks = factorization(self.out_dim, self.dim) - def merge_weight(self, R_weight, org_weight): - R_weight = R_weight.to(org_weight.device, dtype=org_weight.dtype) - if org_weight.dim() == 4: - weight = torch.einsum("oihw, op -> pihw", org_weight, R_weight) - else: - weight = torch.einsum("oi, op -> pi", org_weight, R_weight) - return weight + def calc_updown_kb(self, orig_weight, multiplier): + oft_blocks = self.oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) + oft_blocks = oft_blocks - oft_blocks.transpose(1, 2) # ensure skew-symmetric orthogonal matrix - def get_weight(self, oft_blocks, multiplier=None): - if self.constraint is not None: - constraint = self.constraint.to(oft_blocks.device, dtype=oft_blocks.dtype) + R = oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) + R = R * multiplier + torch.eye(self.block_size, device=orig_weight.device) - block_Q = oft_blocks - oft_blocks.transpose(1, 2) - norm_Q = torch.norm(block_Q.flatten()) - if self.constraint is not None: - new_norm_Q = torch.clamp(norm_Q, max=constraint) - else: - new_norm_Q = norm_Q - block_Q = block_Q * ((new_norm_Q + 1e-8) / (norm_Q + 1e-8)) - m_I = torch.eye(self.num_blocks, device=oft_blocks.device).unsqueeze(0).repeat(self.block_size, 1, 1) - #m_I = torch.eye(self.block_size, device=oft_blocks.device).unsqueeze(0).repeat(self.num_blocks, 1, 1) - block_R = torch.matmul(m_I + block_Q, (m_I - block_Q).inverse()) - - block_R_weighted = multiplier * block_R + (1 - multiplier) * m_I - R = torch.block_diag(*block_R_weighted) - return R - - def calc_updown_kohya(self, orig_weight, multiplier): - R = self.get_weight(self.oft_blocks, multiplier) - merged_weight = self.merge_weight(R, orig_weight) + # This errors out for MultiheadAttention, might need to be handled up-stream + merged_weight = rearrange(orig_weight, '(k n) ... -> k n ...', k=self.num_blocks, n=self.block_size) + merged_weight = torch.einsum( + 'k n m, k n ... -> k m ...', + R, + merged_weight + ) + merged_weight = rearrange(merged_weight, 'k m ... -> (k m) ...') updown = merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight output_shape = orig_weight.shape - orig_weight = orig_weight - return self.finalize_updown(updown, orig_weight, output_shape) - - def calc_updown_kb(self, orig_weight, multiplier): - is_other_linear = type(self.sd_module) in [torch.nn.MultiheadAttention] - - if not is_other_linear: - oft_blocks = self.oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) - - # ensure skew-symmetric matrix - oft_blocks = oft_blocks - oft_blocks.transpose(1, 2) - - R = oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) - R = R * multiplier + torch.eye(self.block_size, device=orig_weight.device) - - merged_weight = rearrange(orig_weight, '(k n) ... -> k n ...', k=self.num_blocks, n=self.block_size) - merged_weight = torch.einsum( - 'k n m, k n ... -> k m ...', - R, - merged_weight - ) - merged_weight = rearrange(merged_weight, 'k m ... -> (k m) ...') - - updown = merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight - output_shape = orig_weight.shape - else: - # FIXME: skip MultiheadAttention for now - #up = self.lin_module.weight.to(orig_weight.device, dtype=orig_weight.dtype) - updown = torch.zeros([orig_weight.shape[1], orig_weight.shape[1]], device=orig_weight.device, dtype=orig_weight.dtype) - output_shape = (orig_weight.shape[1], orig_weight.shape[1]) - return self.finalize_updown(updown, orig_weight, output_shape) def calc_updown(self, orig_weight): - # if alpha is a very small number as in coft, calc_scale will return a almost zero number so we ignore it - #multiplier = self.multiplier() * self.calc_scale() + # if alpha is a very small number as in coft, calc_scale() will return a almost zero number so we ignore it multiplier = self.multiplier() - return self.calc_updown_kb(orig_weight, multiplier) # override to remove the multiplier/scale factor; it's already multiplied in get_weight def finalize_updown(self, updown, orig_weight, output_shape, ex_bias=None): - #return super().finalize_updown(updown, orig_weight, output_shape, ex_bias) - if self.bias is not None: updown = updown.reshape(self.bias.shape) updown += self.bias.to(orig_weight.device, dtype=orig_weight.dtype) From bcfaf3979a9f93e37c418b58c75b02d9570b4354 Mon Sep 17 00:00:00 2001 From: AngelBottomless Date: Thu, 16 Nov 2023 18:43:16 +0900 Subject: [PATCH 056/139] convert/add hypertile options --- modules/hypertile.py | 36 ++++++++++++++++++++++++++++++++++++ modules/processing.py | 21 +++++++++++---------- modules/shared_options.py | 6 ++++++ 3 files changed, 53 insertions(+), 10 deletions(-) diff --git a/modules/hypertile.py b/modules/hypertile.py index 32d8604cc..fee24a8ca 100644 --- a/modules/hypertile.py +++ b/modules/hypertile.py @@ -332,3 +332,39 @@ def split_attention( module.forward = module._original_forward_hypertile del module._original_forward_hypertile del module._split_sizes_hypertile + +def hypertile_context_vae(model:nn.Module, aspect_ratio:float, tile_size:int, opts): + """ + Returns context manager for VAE + """ + enabled = not opts.hypertile_split_vae_attn + swap_size = opts.hypertile_swap_size_vae + max_depth = opts.hypertile_max_depth_vae + tile_size_max = opts.hypertile_max_tile_vae + return split_attention( + model, + aspect_ratio=aspect_ratio, + tile_size=min(tile_size, tile_size_max), + swap_size=swap_size, + disable=not enabled, + max_depth=max_depth, + is_sdxl=False, + ) + +def hypertile_context_unet(model:nn.Module, aspect_ratio:float, tile_size:int, opts, is_sdxl:bool): + """ + Returns context manager for U-Net + """ + enabled = not opts.hypertile_split_unet_attn + swap_size = opts.hypertile_swap_size_unet + max_depth = opts.hypertile_max_depth_unet + tile_size_max = opts.hypertile_max_tile_unet + return split_attention( + model, + aspect_ratio=aspect_ratio, + tile_size=min(tile_size, tile_size_max), + swap_size=swap_size, + disable=not enabled, + max_depth=max_depth, + is_sdxl=is_sdxl, + ) \ No newline at end of file diff --git a/modules/processing.py b/modules/processing.py index e19a09a3c..c622ff337 100644 --- a/modules/processing.py +++ b/modules/processing.py @@ -24,7 +24,7 @@ from modules.shared import opts, cmd_opts, state import modules.shared as shared import modules.paths as paths import modules.face_restoration -from modules.hypertile import split_attention, set_hypertile_seed, largest_tile_size_available +from modules.hypertile import set_hypertile_seed, largest_tile_size_available, hypertile_context_unet, hypertile_context_vae import modules.images as images import modules.styles import modules.sd_models as sd_models @@ -874,7 +874,7 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed: else: if opts.sd_vae_decode_method != 'Full': p.extra_generation_params['VAE Decoder'] = opts.sd_vae_decode_method - with split_attention(p.sd_model.first_stage_model, aspect_ratio = p.width / p.height, tile_size=min(largest_tile_size_available(p.width, p.height), 128), disable=not shared.opts.hypertile_split_vae_attn, is_sdxl=shared.sd_model.is_sdxl): + with hypertile_context_unet(p.sd_model.first_stage_model, aspect_ratio=p.width / p.height, tile_size=largest_tile_size_available(p.width, p.height), is_sdxl=shared.sd_model.is_sdxl, opts=shared.opts): x_samples_ddim = decode_latent_batch(p.sd_model, samples_ddim, target_device=devices.cpu, check_for_nans=True) x_samples_ddim = torch.stack(x_samples_ddim).float() @@ -1144,8 +1144,8 @@ class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing): aspect_ratio = self.width / self.height x = self.rng.next() tile_size = largest_tile_size_available(self.width, self.height) - with split_attention(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=min(tile_size, 128), swap_size=1, disable=not shared.opts.hypertile_split_vae_attn, is_sdxl=shared.sd_model.is_sdxl): - with split_attention(self.sd_model.model, aspect_ratio=aspect_ratio, tile_size=min(tile_size, 256), swap_size=2, disable=not shared.opts.hypertile_split_unet_attn, is_sdxl=shared.sd_model.is_sdxl): + with hypertile_context_vae(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=tile_size, opts=shared.opts): + with hypertile_context_unet(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=tile_size, is_sdxl=shared.sd_model.is_sdxl, opts=shared.opts): devices.torch_gc() samples = self.sampler.sample(self, x, conditioning, unconditional_conditioning, image_conditioning=self.txt2img_image_conditioning(x)) del x @@ -1153,7 +1153,7 @@ class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing): return samples if self.latent_scale_mode is None: - with split_attention(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=min(tile_size, 256), swap_size=1, disable=not shared.opts.hypertile_split_vae_attn, is_sdxl=shared.sd_model.is_sdxl): + with hypertile_context_vae(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=tile_size, opts=shared.opts): decoded_samples = torch.stack(decode_latent_batch(self.sd_model, samples, target_device=devices.cpu, check_for_nans=True)).to(dtype=torch.float32) else: decoded_samples = None @@ -1245,15 +1245,16 @@ class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing): if self.scripts is not None: self.scripts.before_hr(self) tile_size = largest_tile_size_available(target_width, target_height) - with split_attention(self.sd_model.first_stage_model, aspect_ratio=target_width / target_height, tile_size=min(tile_size, 256), swap_size=1, disable=not opts.hypertile_split_vae_attn, is_sdxl=shared.sd_model.is_sdxl): - with split_attention(self.sd_model.model, aspect_ratio=target_width / target_height, tile_size=min(tile_size, 256), swap_size=3, max_depth=1,scale_depth=True, disable=not opts.hypertile_split_unet_attn, is_sdxl=shared.sd_model.is_sdxl): + aspect_ratio = self.width / self.height + with hypertile_context_vae(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=tile_size, opts=shared.opts): + with hypertile_context_unet(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=tile_size, is_sdxl=shared.sd_model.is_sdxl, opts=shared.opts): samples = self.sampler.sample_img2img(self, samples, noise, self.hr_c, self.hr_uc, steps=self.hr_second_pass_steps or self.steps, image_conditioning=image_conditioning) sd_models.apply_token_merging(self.sd_model, self.get_token_merging_ratio()) self.sampler = None devices.torch_gc() - with split_attention(self.sd_model.first_stage_model, aspect_ratio=target_width / target_height, tile_size=min(tile_size, 256), swap_size=1, disable=not opts.hypertile_split_vae_attn, is_sdxl=shared.sd_model.is_sdxl): + with hypertile_context_vae(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=tile_size, opts=shared.opts): decoded_samples = decode_latent_batch(self.sd_model, samples, target_device=devices.cpu, check_for_nans=True) self.is_hr_pass = False @@ -1533,8 +1534,8 @@ class StableDiffusionProcessingImg2Img(StableDiffusionProcessing): x *= self.initial_noise_multiplier aspect_ratio = self.width / self.height tile_size = largest_tile_size_available(self.width, self.height) - with split_attention(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=min(tile_size, 128), swap_size=1, disable=not shared.opts.hypertile_split_vae_attn, is_sdxl=shared.sd_model.is_sdxl): - with split_attention(self.sd_model.model, aspect_ratio=aspect_ratio, tile_size=min(tile_size, 256), swap_size=2, disable=not shared.opts.hypertile_split_unet_attn, is_sdxl=shared.sd_model.is_sdxl): + with hypertile_context_vae(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=tile_size, opts=shared.opts): + with hypertile_context_unet(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=tile_size, is_sdxl=shared.sd_model.is_sdxl, opts=shared.opts): devices.torch_gc() samples = self.sampler.sample_img2img(self, self.init_latent, x, conditioning, unconditional_conditioning, image_conditioning=self.image_conditioning) diff --git a/modules/shared_options.py b/modules/shared_options.py index d96502656..28a489069 100644 --- a/modules/shared_options.py +++ b/modules/shared_options.py @@ -202,6 +202,12 @@ options_templates.update(options_section(('optimizations', "Optimizations"), { "batch_cond_uncond": OptionInfo(True, "Batch cond/uncond").info("do both conditional and unconditional denoising in one batch; uses a bit more VRAM during sampling, but improves speed; previously this was controlled by --always-batch-cond-uncond comandline argument"), "hypertile_split_unet_attn" : OptionInfo(False, "Split attention in Unet with HyperTile").link("Github", "https://github.com/tfernd/HyperTile").info("improves performance; changes behavior, but deterministic"), "hypertile_split_vae_attn": OptionInfo(False, "Split attention in VAE with HyperTile").link("Github", "https://github.com/tfernd/HyperTile").info("improves performance; changes behavior, but deterministic"), + "hypertile_max_depth_vae" : OptionInfo(3, "Max depth for VAE HyperTile hijack", gr.Slider, {"minimum": 0, "maximum": 3, "step": 1}).link("Github", "https://github.com/tfernd/HyperTile"), + "hypertile_max_depth_unet" : OptionInfo(3, "Max depth for Unet HyperTile hijack", gr.Slider, {"minimum": 0, "maximum": 3, "step": 1}).link("Github", "https://github.com/tfernd/HyperTile"), + "hypertile_max_tile_vae" : OptionInfo(128, "Max tile size for VAE HyperTile hijack", gr.Slider, {"minimum": 0, "maximum": 512, "step": 16}).link("Github", "https://github.com/tfernd/HyperTile"), + "hypertile_max_tile_unet" : OptionInfo(256, "Max tile size for Unet HyperTile hijack", gr.Slider, {"minimum": 0, "maximum": 512, "step": 16}).link("Github", "https://github.com/tfernd/HyperTile"), + "hypertile_swap_size_unet": OptionInfo(3, "Swap size for Unet HyperTile hijack", gr.Slider, {"minimum": 0, "maximum": 6, "step": 1}).link("Github", "https://github.com/tfernd/HyperTile"), + "hypertile_swap_size_vae": OptionInfo(3, "Swap size for VAE HyperTile hijack", gr.Slider, {"minimum": 0, "maximum": 6, "step": 1}).link("Github", "https://github.com/tfernd/HyperTile"), })) options_templates.update(options_section(('compatibility', "Compatibility"), { From 472c22cc8a46b825545d5c86bd2745269430d7b0 Mon Sep 17 00:00:00 2001 From: AngelBottomless Date: Thu, 16 Nov 2023 19:03:45 +0900 Subject: [PATCH 057/139] fix ruff - add newline --- modules/hypertile.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/modules/hypertile.py b/modules/hypertile.py index fee24a8ca..86acecdc0 100644 --- a/modules/hypertile.py +++ b/modules/hypertile.py @@ -367,4 +367,4 @@ def hypertile_context_unet(model:nn.Module, aspect_ratio:float, tile_size:int, o disable=not enabled, max_depth=max_depth, is_sdxl=is_sdxl, - ) \ No newline at end of file + ) From 236eb82c3a91960ba5db7b82efbe0f6a9fd7cf24 Mon Sep 17 00:00:00 2001 From: Lucas Daniel Velazquez M <19197331+Luxter77@users.noreply.github.com> Date: Thu, 16 Nov 2023 13:20:33 -0300 Subject: [PATCH 058/139] Adds tqdm handler to logging_config.py for progress bar integration --- modules/logging_config.py | 14 ++++++++++++++ 1 file changed, 14 insertions(+) diff --git a/modules/logging_config.py b/modules/logging_config.py index 7db23d4b6..ce831b5c6 100644 --- a/modules/logging_config.py +++ b/modules/logging_config.py @@ -1,6 +1,19 @@ import os import logging +from tqdm.auto import tqdm + +class TqdmLoggingHandler(logging.Handler): + def __init__(self, level=logging.INFO): + super().__init__(level) + + def emit(self, record): + try: + msg = self.format(record) + tqdm.write(msg) + self.flush() + except Exception: + self.handleError(record) def setup_logging(loglevel): if loglevel is None: @@ -12,5 +25,6 @@ def setup_logging(loglevel): level=log_level, format='%(asctime)s %(levelname)s [%(name)s] %(message)s', datefmt='%Y-%m-%d %H:%M:%S', + handlers=[TqdmLoggingHandler()] ) From cdb60a690dcd35e865eb0caef6c6d8ff64e1b0d5 Mon Sep 17 00:00:00 2001 From: Lucas Daniel Velazquez M <19197331+Luxter77@users.noreply.github.com> Date: Thu, 16 Nov 2023 16:43:59 -0300 Subject: [PATCH 059/139] Take into account tqdm not being installed before first boot for logging --- modules/logging_config.py | 37 ++++++++++++++++++++++++------------- 1 file changed, 24 insertions(+), 13 deletions(-) diff --git a/modules/logging_config.py b/modules/logging_config.py index ce831b5c6..99ed2855b 100644 --- a/modules/logging_config.py +++ b/modules/logging_config.py @@ -1,30 +1,41 @@ import os import logging -from tqdm.auto import tqdm +try: + from tqdm.auto import tqdm -class TqdmLoggingHandler(logging.Handler): - def __init__(self, level=logging.INFO): - super().__init__(level) + class TqdmLoggingHandler(logging.Handler): + def __init__(self, level=logging.INFO): + super().__init__(level) - def emit(self, record): - try: - msg = self.format(record) - tqdm.write(msg) - self.flush() - except Exception: - self.handleError(record) + def emit(self, record): + try: + msg = self.format(record) + tqdm.write(msg) + self.flush() + except Exception: + self.handleError(record) + + TQDM_IMPORTED = True +except ImportError: + # tqdm does not exist before first launch + # I will import once the UI finishes seting up the enviroment and reloads. + TQDM_IMPORTED = False def setup_logging(loglevel): if loglevel is None: loglevel = os.environ.get("SD_WEBUI_LOG_LEVEL") + loghandlers = [] + + if TQDM_IMPORTED: + loghandlers.append(TqdmLoggingHandler()) + if loglevel: log_level = getattr(logging, loglevel.upper(), None) or logging.INFO logging.basicConfig( level=log_level, format='%(asctime)s %(levelname)s [%(name)s] %(message)s', datefmt='%Y-%m-%d %H:%M:%S', - handlers=[TqdmLoggingHandler()] + handlers=[] ) - From 7021cdb1de12be3071ecb67aa8d2e34e7a0ec5ab Mon Sep 17 00:00:00 2001 From: Your Name Date: Thu, 16 Nov 2023 17:53:57 -0300 Subject: [PATCH 060/139] actually adds handler to logging_config.py --- modules/logging_config.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/modules/logging_config.py b/modules/logging_config.py index 99ed2855b..792698756 100644 --- a/modules/logging_config.py +++ b/modules/logging_config.py @@ -37,5 +37,5 @@ def setup_logging(loglevel): level=log_level, format='%(asctime)s %(levelname)s [%(name)s] %(message)s', datefmt='%Y-%m-%d %H:%M:%S', - handlers=[] + handlers=loghandlers ) From c40be2252ab1c8c289562db208c5ac6618bd8545 Mon Sep 17 00:00:00 2001 From: aria1th <35677394+aria1th@users.noreply.github.com> Date: Fri, 17 Nov 2023 09:22:27 +0900 Subject: [PATCH 061/139] Fix critical issue - unet apply --- modules/processing.py | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/modules/processing.py b/modules/processing.py index c622ff337..2fda7f332 100644 --- a/modules/processing.py +++ b/modules/processing.py @@ -874,7 +874,7 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed: else: if opts.sd_vae_decode_method != 'Full': p.extra_generation_params['VAE Decoder'] = opts.sd_vae_decode_method - with hypertile_context_unet(p.sd_model.first_stage_model, aspect_ratio=p.width / p.height, tile_size=largest_tile_size_available(p.width, p.height), is_sdxl=shared.sd_model.is_sdxl, opts=shared.opts): + with hypertile_context_unet(p.sd_model.model, aspect_ratio=p.width / p.height, tile_size=largest_tile_size_available(p.width, p.height), is_sdxl=shared.sd_model.is_sdxl, opts=shared.opts): x_samples_ddim = decode_latent_batch(p.sd_model, samples_ddim, target_device=devices.cpu, check_for_nans=True) x_samples_ddim = torch.stack(x_samples_ddim).float() @@ -1145,7 +1145,7 @@ class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing): x = self.rng.next() tile_size = largest_tile_size_available(self.width, self.height) with hypertile_context_vae(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=tile_size, opts=shared.opts): - with hypertile_context_unet(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=tile_size, is_sdxl=shared.sd_model.is_sdxl, opts=shared.opts): + with hypertile_context_unet(self.sd_model.model, aspect_ratio=aspect_ratio, tile_size=tile_size, is_sdxl=shared.sd_model.is_sdxl, opts=shared.opts): devices.torch_gc() samples = self.sampler.sample(self, x, conditioning, unconditional_conditioning, image_conditioning=self.txt2img_image_conditioning(x)) del x @@ -1247,7 +1247,7 @@ class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing): tile_size = largest_tile_size_available(target_width, target_height) aspect_ratio = self.width / self.height with hypertile_context_vae(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=tile_size, opts=shared.opts): - with hypertile_context_unet(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=tile_size, is_sdxl=shared.sd_model.is_sdxl, opts=shared.opts): + with hypertile_context_unet(self.sd_model.model, aspect_ratio=aspect_ratio, tile_size=tile_size, is_sdxl=shared.sd_model.is_sdxl, opts=shared.opts): samples = self.sampler.sample_img2img(self, samples, noise, self.hr_c, self.hr_uc, steps=self.hr_second_pass_steps or self.steps, image_conditioning=image_conditioning) sd_models.apply_token_merging(self.sd_model, self.get_token_merging_ratio()) @@ -1535,7 +1535,7 @@ class StableDiffusionProcessingImg2Img(StableDiffusionProcessing): aspect_ratio = self.width / self.height tile_size = largest_tile_size_available(self.width, self.height) with hypertile_context_vae(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=tile_size, opts=shared.opts): - with hypertile_context_unet(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=tile_size, is_sdxl=shared.sd_model.is_sdxl, opts=shared.opts): + with hypertile_context_unet(self.sd_model.model, aspect_ratio=aspect_ratio, tile_size=tile_size, is_sdxl=shared.sd_model.is_sdxl, opts=shared.opts): devices.torch_gc() samples = self.sampler.sample_img2img(self, self.init_latent, x, conditioning, unconditional_conditioning, image_conditioning=self.image_conditioning) From c0725ba2d098a6a78610e7d96ee75f63a32d4e52 Mon Sep 17 00:00:00 2001 From: aria1th <35677394+aria1th@users.noreply.github.com> Date: Fri, 17 Nov 2023 09:34:50 +0900 Subject: [PATCH 062/139] Fix inverted option issue I'm pretty sure I was sleepy while implementing this --- modules/hypertile.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/modules/hypertile.py b/modules/hypertile.py index 86acecdc0..3a1468c6b 100644 --- a/modules/hypertile.py +++ b/modules/hypertile.py @@ -337,7 +337,7 @@ def hypertile_context_vae(model:nn.Module, aspect_ratio:float, tile_size:int, op """ Returns context manager for VAE """ - enabled = not opts.hypertile_split_vae_attn + enabled = opts.hypertile_split_vae_attn swap_size = opts.hypertile_swap_size_vae max_depth = opts.hypertile_max_depth_vae tile_size_max = opts.hypertile_max_tile_vae @@ -355,7 +355,7 @@ def hypertile_context_unet(model:nn.Module, aspect_ratio:float, tile_size:int, o """ Returns context manager for U-Net """ - enabled = not opts.hypertile_split_unet_attn + enabled = opts.hypertile_split_unet_attn swap_size = opts.hypertile_swap_size_unet max_depth = opts.hypertile_max_depth_unet tile_size_max = opts.hypertile_max_tile_unet From ffd0f8ddc309688636ac1ac10d82b72ab6b466bf Mon Sep 17 00:00:00 2001 From: aria1th <35677394+aria1th@users.noreply.github.com> Date: Fri, 17 Nov 2023 09:54:33 +0900 Subject: [PATCH 063/139] set empty value for SD XL 3rd layer --- modules/hypertile.py | 1 + 1 file changed, 1 insertion(+) diff --git a/modules/hypertile.py b/modules/hypertile.py index 3a1468c6b..be898fce4 100644 --- a/modules/hypertile.py +++ b/modules/hypertile.py @@ -170,6 +170,7 @@ DEPTH_LAYERS_XL = { "middle_block.1.transformer_blocks.8.attn1", "middle_block.1.transformer_blocks.9.attn1", ], + 3 : [] # TODO - separate layers for SD-XL } From 97431f29feb17ffc96ca95e9b3efec87be9d8b3a Mon Sep 17 00:00:00 2001 From: aria1th <35677394+aria1th@users.noreply.github.com> Date: Fri, 17 Nov 2023 10:05:28 +0900 Subject: [PATCH 064/139] fix double gc and decoding with unet context --- modules/processing.py | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) diff --git a/modules/processing.py b/modules/processing.py index 2fda7f332..36c2be5e5 100644 --- a/modules/processing.py +++ b/modules/processing.py @@ -874,7 +874,7 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed: else: if opts.sd_vae_decode_method != 'Full': p.extra_generation_params['VAE Decoder'] = opts.sd_vae_decode_method - with hypertile_context_unet(p.sd_model.model, aspect_ratio=p.width / p.height, tile_size=largest_tile_size_available(p.width, p.height), is_sdxl=shared.sd_model.is_sdxl, opts=shared.opts): + with hypertile_context_vae(p.sd_model.first_stage_model, aspect_ratio=p.width / p.height, tile_size=largest_tile_size_available(p.width, p.height), opts=shared.opts): x_samples_ddim = decode_latent_batch(p.sd_model, samples_ddim, target_device=devices.cpu, check_for_nans=True) x_samples_ddim = torch.stack(x_samples_ddim).float() @@ -1146,11 +1146,11 @@ class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing): tile_size = largest_tile_size_available(self.width, self.height) with hypertile_context_vae(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=tile_size, opts=shared.opts): with hypertile_context_unet(self.sd_model.model, aspect_ratio=aspect_ratio, tile_size=tile_size, is_sdxl=shared.sd_model.is_sdxl, opts=shared.opts): - devices.torch_gc() samples = self.sampler.sample(self, x, conditioning, unconditional_conditioning, image_conditioning=self.txt2img_image_conditioning(x)) del x if not self.enable_hr: return samples + devices.torch_gc() if self.latent_scale_mode is None: with hypertile_context_vae(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=tile_size, opts=shared.opts): @@ -1536,7 +1536,6 @@ class StableDiffusionProcessingImg2Img(StableDiffusionProcessing): tile_size = largest_tile_size_available(self.width, self.height) with hypertile_context_vae(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=tile_size, opts=shared.opts): with hypertile_context_unet(self.sd_model.model, aspect_ratio=aspect_ratio, tile_size=tile_size, is_sdxl=shared.sd_model.is_sdxl, opts=shared.opts): - devices.torch_gc() samples = self.sampler.sample_img2img(self, self.init_latent, x, conditioning, unconditional_conditioning, image_conditioning=self.image_conditioning) if self.mask is not None: From 4f2a4a361511ca3b8cdd7373f6c7d723583e8fdb Mon Sep 17 00:00:00 2001 From: storyicon Date: Fri, 17 Nov 2023 09:48:18 +0000 Subject: [PATCH 065/139] feat: fix randn found element of type float at pos 2 Signed-off-by: storyicon --- modules/rng.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/modules/rng.py b/modules/rng.py index 9e8ba2ee9..8934d39bf 100644 --- a/modules/rng.py +++ b/modules/rng.py @@ -110,7 +110,7 @@ class ImageRNG: self.is_first = True def first(self): - noise_shape = self.shape if self.seed_resize_from_h <= 0 or self.seed_resize_from_w <= 0 else (self.shape[0], self.seed_resize_from_h // 8, self.seed_resize_from_w // 8) + noise_shape = self.shape if self.seed_resize_from_h <= 0 or self.seed_resize_from_w <= 0 else (self.shape[0], int(self.seed_resize_from_h) // 8, int(self.seed_resize_from_w // 8)) xs = [] From bde439ef67776be126d6a8c569a23d54dbc3e707 Mon Sep 17 00:00:00 2001 From: wfjsw Date: Sun, 19 Nov 2023 00:58:47 -0600 Subject: [PATCH 066/139] use metadata.ini for meta filename --- modules/extensions.py | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/modules/extensions.py b/modules/extensions.py index 5536db3ea..f3988d02e 100644 --- a/modules/extensions.py +++ b/modules/extensions.py @@ -43,13 +43,13 @@ class Extension: @functools.cached_property def metadata(self): - if os.path.isfile(os.path.join(self.path, "sd_webui_metadata.ini")): + if os.path.isfile(os.path.join(self.path, "metadata.ini")): try: config = configparser.ConfigParser() - config.read(os.path.join(self.path, "sd_webui_metadata.ini")) + config.read(os.path.join(self.path, "metadata.ini")) return config except Exception: - errors.report(f"Error reading sd_webui_metadata.ini for extension {self.canonical_name}.", + errors.report(f"Error reading metadata.ini for extension {self.canonical_name}.", exc_info=True) return None @@ -177,14 +177,14 @@ def list_extensions(): canonical_name = extension_dirname requires = None - if os.path.isfile(os.path.join(path, "sd_webui_metadata.ini")): + if os.path.isfile(os.path.join(path, "metadata.ini")): try: config = configparser.ConfigParser() - config.read(os.path.join(path, "sd_webui_metadata.ini")) + config.read(os.path.join(path, "metadata.ini")) canonical_name = config.get("Extension", "Name", fallback=canonical_name) requires = config.get("Extension", "Requires", fallback=None) except Exception: - errors.report(f"Error reading sd_webui_metadata.ini for extension {extension_dirname}. " + errors.report(f"Error reading metadata.ini for extension {extension_dirname}. " f"Will load regardless.", exc_info=True) canonical_name = canonical_name.lower().strip() From dea5e43c8359b663d5599efc99278c258747db61 Mon Sep 17 00:00:00 2001 From: w-e-w <40751091+w-e-w@users.noreply.github.com> Date: Sat, 18 Nov 2023 04:23:03 +0900 Subject: [PATCH 067/139] Option to show batch img2img results in UI shared.opts.img2img_batch_show_results_limit limit the number of images return to the UI for batch img2img default limit 32 0 no images are shown -1 unlimited, all images are shown --- modules/img2img.py | 24 ++++++++++++++++++++---- modules/shared_options.py | 1 + 2 files changed, 21 insertions(+), 4 deletions(-) diff --git a/modules/img2img.py b/modules/img2img.py index 52cb577a6..c583290a0 100644 --- a/modules/img2img.py +++ b/modules/img2img.py @@ -44,6 +44,8 @@ def process_batch(p, input_dir, output_dir, inpaint_mask_dir, args, to_scale=Fal steps = p.steps override_settings = p.override_settings sd_model_checkpoint_override = get_closet_checkpoint_match(override_settings.get("sd_model_checkpoint", None)) + batch_results = None + discard_further_results = False for i, image in enumerate(images): state.job = f"{i+1} out of {len(images)}" if state.skipped: @@ -127,7 +129,21 @@ def process_batch(p, input_dir, output_dir, inpaint_mask_dir, args, to_scale=Fal if proc is None: p.override_settings.pop('save_images_replace_action', None) - process_images(p) + proc = process_images(p) + + if not discard_further_results and proc: + if batch_results: + batch_results.images.extend(proc.images) + batch_results.infotexts.extend(proc.infotexts) + else: + batch_results = proc + + if 0 <= shared.opts.img2img_batch_show_results_limit < len(batch_results.images): + discard_further_results = True + batch_results.images = batch_results.images[:int(shared.opts.img2img_batch_show_results_limit)] + batch_results.infotexts = batch_results.infotexts[:int(shared.opts.img2img_batch_show_results_limit)] + + return batch_results def img2img(id_task: str, mode: int, prompt: str, negative_prompt: str, prompt_styles, init_img, sketch, init_img_with_mask, inpaint_color_sketch, inpaint_color_sketch_orig, init_img_inpaint, init_mask_inpaint, steps: int, sampler_name: str, mask_blur: int, mask_alpha: float, inpainting_fill: int, n_iter: int, batch_size: int, cfg_scale: float, image_cfg_scale: float, denoising_strength: float, selected_scale_tab: int, height: int, width: int, scale_by: float, resize_mode: int, inpaint_full_res: bool, inpaint_full_res_padding: int, inpainting_mask_invert: int, img2img_batch_input_dir: str, img2img_batch_output_dir: str, img2img_batch_inpaint_mask_dir: str, override_settings_texts, img2img_batch_use_png_info: bool, img2img_batch_png_info_props: list, img2img_batch_png_info_dir: str, request: gr.Request, *args): @@ -212,10 +228,10 @@ def img2img(id_task: str, mode: int, prompt: str, negative_prompt: str, prompt_s with closing(p): if is_batch: assert not shared.cmd_opts.hide_ui_dir_config, "Launched with --hide-ui-dir-config, batch img2img disabled" + processed = process_batch(p, img2img_batch_input_dir, img2img_batch_output_dir, img2img_batch_inpaint_mask_dir, args, to_scale=selected_scale_tab == 1, scale_by=scale_by, use_png_info=img2img_batch_use_png_info, png_info_props=img2img_batch_png_info_props, png_info_dir=img2img_batch_png_info_dir) - process_batch(p, img2img_batch_input_dir, img2img_batch_output_dir, img2img_batch_inpaint_mask_dir, args, to_scale=selected_scale_tab == 1, scale_by=scale_by, use_png_info=img2img_batch_use_png_info, png_info_props=img2img_batch_png_info_props, png_info_dir=img2img_batch_png_info_dir) - - processed = Processed(p, [], p.seed, "") + if processed is None: + processed = Processed(p, [], p.seed, "") else: processed = modules.scripts.scripts_img2img.run(p, *args) if processed is None: diff --git a/modules/shared_options.py b/modules/shared_options.py index d40db5306..1ee8c7ad1 100644 --- a/modules/shared_options.py +++ b/modules/shared_options.py @@ -189,6 +189,7 @@ options_templates.update(options_section(('img2img', "img2img"), { "img2img_inpaint_sketch_default_brush_color": OptionInfo("#ffffff", "Inpaint sketch initial brush color", ui_components.FormColorPicker, {}).info("default brush color of img2img inpaint sketch").needs_reload_ui(), "return_mask": OptionInfo(False, "For inpainting, include the greyscale mask in results for web"), "return_mask_composite": OptionInfo(False, "For inpainting, include masked composite in results for web"), + "img2img_batch_show_results_limit": OptionInfo(32, "Show the first N batch img2img results in UI", gr.Slider, {"minimum": -1, "maximum": 1000, "step": 1}).info('0: disable, -1: show all images. Too many images can cause lag'), })) options_templates.update(options_section(('optimizations', "Optimizations"), { From 6d337bf23dae990e7b6717da4d5f2e54f212685c Mon Sep 17 00:00:00 2001 From: w-e-w <40751091+w-e-w@users.noreply.github.com> Date: Mon, 20 Nov 2023 01:38:31 +0900 Subject: [PATCH 068/139] save sysinfo as .json GitHub now allows uploading of .json files in issues --- modules/launch_utils.py | 2 +- modules/ui.py | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/modules/launch_utils.py b/modules/launch_utils.py index 8cdbafa50..264ec9ca6 100644 --- a/modules/launch_utils.py +++ b/modules/launch_utils.py @@ -441,7 +441,7 @@ def dump_sysinfo(): import datetime text = sysinfo.get() - filename = f"sysinfo-{datetime.datetime.utcnow().strftime('%Y-%m-%d-%H-%M')}.txt" + filename = f"sysinfo-{datetime.datetime.utcnow().strftime('%Y-%m-%d-%H-%M')}.json" with open(filename, "w", encoding="utf8") as file: file.write(text) diff --git a/modules/ui.py b/modules/ui.py index ba0d8542b..b82f3c5e8 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -1308,7 +1308,7 @@ def setup_ui_api(app): from fastapi.responses import PlainTextResponse text = sysinfo.get() - filename = f"sysinfo-{datetime.datetime.utcnow().strftime('%Y-%m-%d-%H-%M')}.txt" + filename = f"sysinfo-{datetime.datetime.utcnow().strftime('%Y-%m-%d-%H-%M')}.json" return PlainTextResponse(text, headers={'Content-Disposition': f'{"attachment" if attachment else "inline"}; filename="{filename}"'}) From 9b471436b2226458a767077707ea102e331b5d78 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Mon, 20 Nov 2023 14:47:09 +0300 Subject: [PATCH 069/139] rework extensions metadata: use custom sorter that doesn't mess the order as much and ignores cyclic errors, use classes with named fields instead of dictionaries, eliminate some duplicated code --- modules/extensions.py | 132 +++++++++++++++++---------------- modules/scripts.py | 165 +++++++++++++++++++----------------------- 2 files changed, 146 insertions(+), 151 deletions(-) diff --git a/modules/extensions.py b/modules/extensions.py index f3988d02e..1899cd529 100644 --- a/modules/extensions.py +++ b/modules/extensions.py @@ -1,5 +1,6 @@ +from __future__ import annotations + import configparser -import functools import os import threading import re @@ -8,7 +9,6 @@ from modules import shared, errors, cache, scripts from modules.gitpython_hack import Repo from modules.paths_internal import extensions_dir, extensions_builtin_dir, script_path # noqa: F401 -extensions = [] os.makedirs(extensions_dir, exist_ok=True) @@ -22,13 +22,56 @@ def active(): return [x for x in extensions if x.enabled] +class ExtensionMetadata: + filename = "metadata.ini" + config: configparser.ConfigParser + canonical_name: str + requires: list + + def __init__(self, path, canonical_name): + self.config = configparser.ConfigParser() + + filepath = os.path.join(path, self.filename) + if os.path.isfile(filepath): + try: + self.config.read(filepath) + except Exception: + errors.report(f"Error reading {self.filename} for extension {canonical_name}.", exc_info=True) + + self.canonical_name = self.config.get("Extension", "Name", fallback=canonical_name) + self.canonical_name = canonical_name.lower().strip() + + self.requires = self.get_script_requirements("Requires", "Extension") + + def get_script_requirements(self, field, section, extra_section=None): + """reads a list of requirements from the config; field is the name of the field in the ini file, + like Requires or Before, and section is the name of the [section] in the ini file; additionally, + reads more requirements from [extra_section] if specified.""" + + x = self.config.get(section, field, fallback='') + + if extra_section: + x = x + ', ' + self.config.get(extra_section, field, fallback='') + + return self.parse_list(x.lower()) + + def parse_list(self, text): + """converts a line from config ("ext1 ext2, ext3 ") into a python list (["ext1", "ext2", "ext3"])""" + + if not text: + return [] + + # both "," and " " are accepted as separator + return [x for x in re.split(r"[,\s]+", text.strip()) if x] + + class Extension: lock = threading.Lock() cached_fields = ['remote', 'commit_date', 'branch', 'commit_hash', 'version'] + metadata: ExtensionMetadata - def __init__(self, name, path, enabled=True, is_builtin=False, canonical_name=None): + def __init__(self, name, path, enabled=True, is_builtin=False, metadata=None): self.name = name - self.canonical_name = canonical_name or name.lower() self.path = path self.enabled = enabled self.status = '' @@ -40,18 +83,8 @@ class Extension: self.branch = None self.remote = None self.have_info_from_repo = False - - @functools.cached_property - def metadata(self): - if os.path.isfile(os.path.join(self.path, "metadata.ini")): - try: - config = configparser.ConfigParser() - config.read(os.path.join(self.path, "metadata.ini")) - return config - except Exception: - errors.report(f"Error reading metadata.ini for extension {self.canonical_name}.", - exc_info=True) - return None + self.metadata = metadata if metadata else ExtensionMetadata(self.path, name.lower()) + self.canonical_name = metadata.canonical_name def to_dict(self): return {x: getattr(self, x) for x in self.cached_fields} @@ -162,7 +195,7 @@ def list_extensions(): elif shared.opts.disable_all_extensions == "extra": print("*** \"Disable all extensions\" option was set, will only load built-in extensions ***") - extension_dependency_map = {} + loaded_extensions = {} # scan through extensions directory and load metadata for dirname in [extensions_builtin_dir, extensions_dir]: @@ -175,55 +208,30 @@ def list_extensions(): continue canonical_name = extension_dirname - requires = None - - if os.path.isfile(os.path.join(path, "metadata.ini")): - try: - config = configparser.ConfigParser() - config.read(os.path.join(path, "metadata.ini")) - canonical_name = config.get("Extension", "Name", fallback=canonical_name) - requires = config.get("Extension", "Requires", fallback=None) - except Exception: - errors.report(f"Error reading metadata.ini for extension {extension_dirname}. " - f"Will load regardless.", exc_info=True) - - canonical_name = canonical_name.lower().strip() + metadata = ExtensionMetadata(path, canonical_name) # check for duplicated canonical names - if canonical_name in extension_dependency_map: - errors.report(f"Duplicate canonical name \"{canonical_name}\" found in extensions " - f"\"{extension_dirname}\" and \"{extension_dependency_map[canonical_name]['dirname']}\". " - f"The current loading extension will be discarded.", exc_info=False) + already_loaded_extension = loaded_extensions.get(metadata.canonical_name) + if already_loaded_extension is not None: + errors.report(f'Duplicate canonical name "{canonical_name}" found in extensions "{extension_dirname}" and "{already_loaded_extension.name}". Former will be discarded.', exc_info=False) continue - # both "," and " " are accepted as separator - requires = list(filter(None, re.split(r"[,\s]+", requires.lower()))) if requires else [] - - extension_dependency_map[canonical_name] = { - "dirname": extension_dirname, - "path": path, - "requires": requires, - } + is_builtin = dirname == extensions_builtin_dir + extension = Extension(name=extension_dirname, path=path, enabled=extension_dirname not in shared.opts.disabled_extensions, is_builtin=is_builtin, metadata=metadata) + extensions.append(extension) + loaded_extensions[canonical_name] = extension # check for requirements - for (_, extension_data) in extension_dependency_map.items(): - dirname, path, requires = extension_data['dirname'], extension_data['path'], extension_data['requires'] - requirement_met = True - for req in requires: - if req not in extension_dependency_map: - errors.report(f"Extension \"{dirname}\" requires \"{req}\" which is not installed. " - f"The current loading extension will be discarded.", exc_info=False) - requirement_met = False - break - dep_dirname = extension_dependency_map[req]['dirname'] - if dep_dirname in shared.opts.disabled_extensions: - errors.report(f"Extension \"{dirname}\" requires \"{dep_dirname}\" which is disabled. " - f"The current loading extension will be discarded.", exc_info=False) - requirement_met = False - break + for extension in extensions: + for req in extension.metadata.requires: + required_extension = loaded_extensions.get(req) + if required_extension is None: + errors.report(f'Extension "{extension.name}" requires "{req}" which is not installed.', exc_info=False) + continue - is_builtin = dirname == extensions_builtin_dir - extension = Extension(name=dirname, path=path, - enabled=dirname not in shared.opts.disabled_extensions and requirement_met, - is_builtin=is_builtin) - extensions.append(extension) + if not extension.enabled: + errors.report(f'Extension "{extension.name}" requires "{required_extension.name}" which is disabled.', exc_info=False) + continue + + +extensions: list[Extension] = [] diff --git a/modules/scripts.py b/modules/scripts.py index b1f4504a5..b0689a23d 100644 --- a/modules/scripts.py +++ b/modules/scripts.py @@ -2,7 +2,6 @@ import os import re import sys import inspect -from graphlib import TopologicalSorter, CycleError from collections import namedtuple from dataclasses import dataclass @@ -312,27 +311,57 @@ scripts_data = [] postprocessing_scripts_data = [] ScriptClassData = namedtuple("ScriptClassData", ["script_class", "path", "basedir", "module"]) +def topological_sort(dependencies): + """Accepts a dictionary mapping name to its dependencies, returns a list of names ordered according to dependencies. + Ignores errors relating to missing dependeencies or circular dependencies + """ + + visited = {} + result = [] + + def inner(name): + visited[name] = True + + for dep in dependencies.get(name, []): + if dep in dependencies and dep not in visited: + inner(dep) + + result.append(name) + + for depname in dependencies: + if depname not in visited: + inner(depname) + + return result + + +@dataclass +class ScriptWithDependencies: + script_canonical_name: str + file: ScriptFile + requires: list + load_before: list + load_after: list + def list_scripts(scriptdirname, extension, *, include_extensions=True): - scripts_list = [] - script_dependency_map = {} + scripts = {} + + loaded_extensions = {ext.canonical_name: ext for ext in extensions.active()} + loaded_extensions_scripts = {ext.canonical_name: [] for ext in extensions.active()} # build script dependency map - root_script_basedir = os.path.join(paths.script_path, scriptdirname) if os.path.exists(root_script_basedir): for filename in sorted(os.listdir(root_script_basedir)): if not os.path.isfile(os.path.join(root_script_basedir, filename)): continue - script_dependency_map[filename] = { - "extension": None, - "extension_dirname": None, - "script_file": ScriptFile(paths.script_path, filename, os.path.join(root_script_basedir, filename)), - "requires": [], - "load_before": [], - "load_after": [], - } + if os.path.splitext(filename)[1].lower() != extension: + continue + + script_file = ScriptFile(paths.script_path, filename, os.path.join(root_script_basedir, filename)) + scripts[filename] = ScriptWithDependencies(filename, script_file, [], [], []) if include_extensions: for ext in extensions.active(): @@ -341,96 +370,54 @@ def list_scripts(scriptdirname, extension, *, include_extensions=True): if not os.path.isfile(extension_script.path): continue - script_canonical_name = ext.canonical_name + "/" + extension_script.filename - if ext.is_builtin: - script_canonical_name = "builtin/" + script_canonical_name + script_canonical_name = ("builtin/" if ext.is_builtin else "") + ext.canonical_name + "/" + extension_script.filename relative_path = scriptdirname + "/" + extension_script.filename - requires = '' - load_before = '' - load_after = '' + script = ScriptWithDependencies( + script_canonical_name=script_canonical_name, + file=extension_script, + requires=ext.metadata.get_script_requirements("Requires", relative_path, scriptdirname), + load_before=ext.metadata.get_script_requirements("Before", relative_path, scriptdirname), + load_after=ext.metadata.get_script_requirements("After", relative_path, scriptdirname), + ) - if ext.metadata is not None: - requires = ext.metadata.get(relative_path, "Requires", fallback='') - load_before = ext.metadata.get(relative_path, "Before", fallback='') - load_after = ext.metadata.get(relative_path, "After", fallback='') + scripts[script_canonical_name] = script + loaded_extensions_scripts[ext.canonical_name].append(script) - # propagate directory level metadata - requires = requires + ',' + ext.metadata.get(scriptdirname, "Requires", fallback='') - load_before = load_before + ',' + ext.metadata.get(scriptdirname, "Before", fallback='') - load_after = load_after + ',' + ext.metadata.get(scriptdirname, "After", fallback='') - - requires = list(filter(None, re.split(r"[,\s]+", requires.lower()))) if requires else [] - load_after = list(filter(None, re.split(r"[,\s]+", load_after.lower()))) if load_after else [] - load_before = list(filter(None, re.split(r"[,\s]+", load_before.lower()))) if load_before else [] - - script_dependency_map[script_canonical_name] = { - "extension": ext.canonical_name, - "extension_dirname": ext.name, - "script_file": extension_script, - "requires": requires, - "load_before": load_before, - "load_after": load_after, - } - - # resolve dependencies - - loaded_extensions = set() - for ext in extensions.active(): - loaded_extensions.add(ext.canonical_name) - - for script_canonical_name, script_data in script_dependency_map.items(): + for script_canonical_name, script in scripts.items(): # load before requires inverse dependency # in this case, append the script name into the load_after list of the specified script - for load_before_script in script_data['load_before']: + for load_before in script.load_before: # if this requires an individual script to be loaded before - if load_before_script in script_dependency_map: - script_dependency_map[load_before_script]['load_after'].append(script_canonical_name) - elif load_before_script in loaded_extensions: - for _, script_data2 in script_dependency_map.items(): - if script_data2['extension'] == load_before_script: - script_data2['load_after'].append(script_canonical_name) - break + other_script = scripts.get(load_before) + if other_script: + other_script.load_after.append(script_canonical_name) - # resolve extension name in load_after lists - for load_after_script in list(script_data['load_after']): - if load_after_script not in script_dependency_map and load_after_script in loaded_extensions: - script_data['load_after'].remove(load_after_script) - for script_canonical_name2, script_data2 in script_dependency_map.items(): - if script_data2['extension'] == load_after_script: - script_data['load_after'].append(script_canonical_name2) - break + # if this requires an extension + other_extension_scripts = loaded_extensions_scripts.get(load_before) + if other_extension_scripts: + for other_script in other_extension_scripts: + other_script.load_after.append(script_canonical_name) - # build the DAG - sorter = TopologicalSorter() - for script_canonical_name, script_data in script_dependency_map.items(): - requirement_met = True - for required_script in script_data['requires']: - # if this requires an individual script to be loaded - if required_script not in script_dependency_map and required_script not in loaded_extensions: - errors.report(f"Script \"{script_canonical_name}\" " - f"requires \"{required_script}\" to " - f"be loaded, but it is not. Skipping.", - exc_info=False) - requirement_met = False - break - if not requirement_met: - continue + # if After mentions an extension, remove it and instead add all of its scripts + for load_after in list(script.load_after): + if load_after not in scripts and load_after in loaded_extensions_scripts: + script.load_after.remove(load_after) - sorter.add(script_canonical_name, *script_data['load_after']) + for other_script in loaded_extensions_scripts.get(load_after, []): + script.load_after.append(other_script.script_canonical_name) - # sort the scripts - try: - ordered_script = sorter.static_order() - except CycleError: - errors.report("Cycle detected in script dependencies. Scripts will load in ascending order.", exc_info=True) - ordered_script = script_dependency_map.keys() + dependencies = {} - for script_canonical_name in ordered_script: - script_data = script_dependency_map[script_canonical_name] - scripts_list.append(script_data['script_file']) + for script_canonical_name, script in scripts.items(): + for required_script in script.requires: + if required_script not in scripts and required_script not in loaded_extensions: + errors.report(f'Script "{script_canonical_name}" requires "{required_script}" to be loaded, but it is not.', exc_info=False) - scripts_list = [x for x in scripts_list if os.path.splitext(x.path)[1].lower() == extension and os.path.isfile(x.path)] + dependencies[script_canonical_name] = script.load_after + + ordered_scripts = topological_sort(dependencies) + scripts_list = [scripts[script_canonical_name].file for script_canonical_name in ordered_scripts] return scripts_list From 314ae1535ea172fcdb0f5b3b2eecc5d4ce9112b5 Mon Sep 17 00:00:00 2001 From: Tom Haelbich Date: Mon, 20 Nov 2023 16:19:54 +0100 Subject: [PATCH 070/139] added option for default behavior of dir buttons --- modules/shared_options.py | 1 + modules/ui_extra_networks.py | 9 +++++++-- 2 files changed, 8 insertions(+), 2 deletions(-) diff --git a/modules/shared_options.py b/modules/shared_options.py index 00b273faa..1d2dca797 100644 --- a/modules/shared_options.py +++ b/modules/shared_options.py @@ -224,6 +224,7 @@ options_templates.update(options_section(('interrogate', "Interrogate"), { options_templates.update(options_section(('extra_networks', "Extra Networks"), { "extra_networks_show_hidden_directories": OptionInfo(True, "Show hidden directories").info("directory is hidden if its name starts with \".\"."), + "extra_networks_dir_button_function": OptionInfo(False, "Add a '/' to the beginning of directory buttons").info("Buttons will display the contents of the selected directory without acting as a search filter."), "extra_networks_hidden_models": OptionInfo("When searched", "Show cards for models in hidden directories", gr.Radio, {"choices": ["Always", "When searched", "Never"]}).info('"When searched" option will only show the item when the search string has 4 characters or more'), "extra_networks_default_multiplier": OptionInfo(1.0, "Default multiplier for extra networks", gr.Slider, {"minimum": 0.0, "maximum": 2.0, "step": 0.01}), "extra_networks_card_width": OptionInfo(0, "Card width for Extra Networks").info("in pixels"), diff --git a/modules/ui_extra_networks.py b/modules/ui_extra_networks.py index bd6732856..27a37295f 100644 --- a/modules/ui_extra_networks.py +++ b/modules/ui_extra_networks.py @@ -138,8 +138,13 @@ class ExtraNetworksPage: continue subdir = os.path.abspath(x)[len(parentdir):].replace("\\", "/") - if not subdir.startswith("/"): - subdir = "/" + subdir + + if shared.opts.extra_networks_dir_button_function: + if not subdir.startswith("/"): + subdir = "/" + subdir + else: + while subdir.startswith("/"): + subdir = subdir[1:] is_empty = len(os.listdir(x)) == 0 if not is_empty and not subdir.endswith("/"): From 58c19545c83fa6925c9ce2216ee64964eb5129ce Mon Sep 17 00:00:00 2001 From: hidenorly Date: Tue, 21 Nov 2023 01:13:53 +0900 Subject: [PATCH 071/139] Add FP32 fallback support on sd_vae_approx This tries to execute interpolate with FP32 if it failed. Background is that on some environment such as Mx chip MacOS devices, we get error as follows: ``` "torch/nn/functional.py", line 3931, in interpolate return torch._C._nn.upsample_nearest2d(input, output_size, scale_factors) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ RuntimeError: "upsample_nearest2d_channels_last" not implemented for 'Half' ``` In this case, ```--no-half``` doesn't help to solve. Therefore this commits add the FP32 fallback execution to solve it. Note that the submodule may require additional modifications. The following is the example modification on the other submodule. ```repositories/stable-diffusion-stability-ai/ldm/modules/diffusionmodules/openaimodel.py class Upsample(nn.Module): ..snip.. def forward(self, x): assert x.shape[1] == self.channels if self.dims == 3: x = F.interpolate( x, (x.shape[2], x.shape[3] * 2, x.shape[4] * 2), mode="nearest" ) else: try: x = F.interpolate(x, scale_factor=2, mode="nearest") except: x = F.interpolate(x.to(th.float32), scale_factor=2, mode="nearest").to(x.dtype) if self.use_conv: x = self.conv(x) return x ..snip.. ``` You can see the FP32 fallback execution as same as sd_vae_approx.py. --- modules/sd_vae_approx.py | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-) diff --git a/modules/sd_vae_approx.py b/modules/sd_vae_approx.py index 3965e223e..8370493f9 100644 --- a/modules/sd_vae_approx.py +++ b/modules/sd_vae_approx.py @@ -21,7 +21,13 @@ class VAEApprox(nn.Module): def forward(self, x): extra = 11 - x = nn.functional.interpolate(x, (x.shape[2] * 2, x.shape[3] * 2)) + try: + x = nn.functional.interpolate(x, (x.shape[2] * 2, x.shape[3] * 2)) + except RuntimeError as e: + if "not implemented for" in str(e) and "Half" in str(e): + x = nn.functional.interpolate(x.to(torch.float32), (x.shape[2] * 2, x.shape[3] * 2)).to(x.dtype) + else: + print(f"An unexpected RuntimeError occurred: {str(e)}") x = nn.functional.pad(x, (extra, extra, extra, extra)) for layer in [self.conv1, self.conv2, self.conv3, self.conv4, self.conv5, self.conv6, self.conv7, self.conv8, ]: From 8aa51f682c17d85f4585b9471860224568d25e95 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Tue, 21 Nov 2023 08:32:00 +0300 Subject: [PATCH 072/139] fix [Bug]: (Dev Branch) Placing "Dimensions" first in "ui_reorder_list" prevents start #14047 --- modules/ui.py | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/modules/ui.py b/modules/ui.py index b82f3c5e8..08e0ad775 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -635,12 +635,6 @@ def create_ui(): scale_by.release(**on_change_args) button_update_resize_to.click(**on_change_args) - # the code below is meant to update the resolution label after the image in the image selection UI has changed. - # as it is now the event keeps firing continuously for inpaint edits, which ruins the page with constant requests. - # I assume this must be a gradio bug and for now we'll just do it for non-inpaint inputs. - for component in [init_img, sketch]: - component.change(fn=lambda: None, _js="updateImg2imgResizeToTextAfterChangingImage", inputs=[], outputs=[], show_progress=False) - tab_scale_to.select(fn=lambda: 0, inputs=[], outputs=[selected_scale_tab]) tab_scale_by.select(fn=lambda: 1, inputs=[], outputs=[selected_scale_tab]) @@ -701,6 +695,12 @@ def create_ui(): if category not in {"accordions"}: scripts.scripts_img2img.setup_ui_for_section(category) + # the code below is meant to update the resolution label after the image in the image selection UI has changed. + # as it is now the event keeps firing continuously for inpaint edits, which ruins the page with constant requests. + # I assume this must be a gradio bug and for now we'll just do it for non-inpaint inputs. + for component in [init_img, sketch]: + component.change(fn=lambda: None, _js="updateImg2imgResizeToTextAfterChangingImage", inputs=[], outputs=[], show_progress=False) + def select_img2img_tab(tab): return gr.update(visible=tab in [2, 3, 4]), gr.update(visible=tab == 3), From 8fe1e195228162a4510925de05015f361efa1087 Mon Sep 17 00:00:00 2001 From: Aarni Koskela Date: Wed, 22 Nov 2023 18:01:34 +0200 Subject: [PATCH 073/139] Update ruff to 0.1.6 --- .github/workflows/on_pull_request.yaml | 2 +- pyproject.toml | 1 + 2 files changed, 2 insertions(+), 1 deletion(-) diff --git a/.github/workflows/on_pull_request.yaml b/.github/workflows/on_pull_request.yaml index 78e608ee9..9e44c806a 100644 --- a/.github/workflows/on_pull_request.yaml +++ b/.github/workflows/on_pull_request.yaml @@ -20,7 +20,7 @@ jobs: # not to have GHA download an (at the time of writing) 4 GB cache # of PyTorch and other dependencies. - name: Install Ruff - run: pip install ruff==0.0.272 + run: pip install ruff==0.1.6 - name: Run Ruff run: ruff . lint-js: diff --git a/pyproject.toml b/pyproject.toml index 80541a8f3..d03036e7d 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -16,6 +16,7 @@ exclude = [ ignore = [ "E501", # Line too long + "E721", # Do not compare types, use `isinstance` "E731", # Do not assign a `lambda` expression, use a `def` "I001", # Import block is un-sorted or un-formatted From 066afda2f6f650fe108d285a239d08d59d92590d Mon Sep 17 00:00:00 2001 From: Aarni Koskela Date: Wed, 22 Nov 2023 18:02:39 +0200 Subject: [PATCH 074/139] Simplify restart_sampler (suggested by ruff) --- modules/sd_samplers_extra.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/modules/sd_samplers_extra.py b/modules/sd_samplers_extra.py index 1b981ca80..72fd0aa5e 100644 --- a/modules/sd_samplers_extra.py +++ b/modules/sd_samplers_extra.py @@ -60,7 +60,7 @@ def restart_sampler(model, x, sigmas, extra_args=None, callback=None, disable=No sigma_restart = get_sigmas_karras(restart_steps, sigmas[min_idx].item(), sigmas[max_idx].item(), device=sigmas.device)[:-1] while restart_times > 0: restart_times -= 1 - step_list.extend([(old_sigma, new_sigma) for (old_sigma, new_sigma) in zip(sigma_restart[:-1], sigma_restart[1:])]) + step_list.extend(zip(sigma_restart[:-1], sigma_restart[1:])) last_sigma = None for old_sigma, new_sigma in tqdm.tqdm(step_list, disable=disable): From ac2a981c4f30d77cdb674948fe0e2aa7264a93e1 Mon Sep 17 00:00:00 2001 From: wfjsw Date: Wed, 22 Nov 2023 22:40:24 -0600 Subject: [PATCH 075/139] use extension name for determining an extension is installed in the index --- modules/ui_extensions.py | 14 ++++++++------ 1 file changed, 8 insertions(+), 6 deletions(-) diff --git a/modules/ui_extensions.py b/modules/ui_extensions.py index c0a73b573..b67088811 100644 --- a/modules/ui_extensions.py +++ b/modules/ui_extensions.py @@ -335,6 +335,11 @@ def normalize_git_url(url): return url +def get_extension_dirname_from_url(url): + *parts, last_part = url.split('/') + return normalize_git_url(last_part) + + def install_extension_from_url(dirname, url, branch_name=None): check_access() @@ -346,10 +351,7 @@ def install_extension_from_url(dirname, url, branch_name=None): assert url, 'No URL specified' if dirname is None or dirname == "": - *parts, last_part = url.split('/') - last_part = normalize_git_url(last_part) - - dirname = last_part + dirname = get_extension_dirname_from_url(url) target_dir = os.path.join(extensions.extensions_dir, dirname) assert not os.path.exists(target_dir), f'Extension directory already exists: {target_dir}' @@ -449,7 +451,7 @@ def get_date(info: dict, key): def refresh_available_extensions_from_data(hide_tags, sort_column, filter_text=""): extlist = available_extensions["extensions"] - installed_extension_urls = {normalize_git_url(extension.remote): extension.name for extension in extensions.extensions} + installed_extensions = {extension.name for extension in extensions.extensions} tags = available_extensions.get("tags", {}) tags_to_hide = set(hide_tags) @@ -482,7 +484,7 @@ def refresh_available_extensions_from_data(hide_tags, sort_column, filter_text=" if url is None: continue - existing = installed_extension_urls.get(normalize_git_url(url), None) + existing = get_extension_dirname_from_url(url) in installed_extensions extension_tags = extension_tags + ["installed"] if existing else extension_tags if any(x for x in extension_tags if x in tags_to_hide): From 86b99b1e98fcdd6e7e5f6017071944364e01e6ad Mon Sep 17 00:00:00 2001 From: Jabasukuriputo Wang Date: Fri, 24 Nov 2023 11:28:54 -0600 Subject: [PATCH 076/139] Move exception_records related methods to errors.py --- modules/errors.py | 18 ++++++++++++++++-- modules/sysinfo.py | 17 +---------------- 2 files changed, 17 insertions(+), 18 deletions(-) diff --git a/modules/errors.py b/modules/errors.py index 192cd8ffd..ac9f1ee5e 100644 --- a/modules/errors.py +++ b/modules/errors.py @@ -6,6 +6,21 @@ import traceback exception_records = [] +def format_traceback(tb): + return [[f"{x.filename}, line {x.lineno}, {x.name}", x.line] for x in traceback.extract_tb(tb)] + + +def format_exception(e, tb): + return {"exception": str(e), "traceback": format_traceback(tb)} + + +def get_exceptions(): + try: + return list(reversed(exception_records)) + except Exception as e: + return str(e) + + def record_exception(): _, e, tb = sys.exc_info() if e is None: @@ -14,8 +29,7 @@ def record_exception(): if exception_records and exception_records[-1] == e: return - from modules import sysinfo - exception_records.append(sysinfo.format_exception(e, tb)) + exception_records.append(format_exception(e, tb)) if len(exception_records) > 5: exception_records.pop(0) diff --git a/modules/sysinfo.py b/modules/sysinfo.py index 7d906e1fe..226b204d9 100644 --- a/modules/sysinfo.py +++ b/modules/sysinfo.py @@ -85,7 +85,7 @@ def get_dict(): "Checksum": checksum_token, "Commandline": sys.argv, "Torch env info": get_torch_sysinfo(), - "Exceptions": get_exceptions(), + "Exceptions": errors.get_exceptions(), "CPU": { "model": platform.processor(), "count logical": psutil.cpu_count(logical=True), @@ -105,21 +105,6 @@ def get_dict(): return res -def format_traceback(tb): - return [[f"{x.filename}, line {x.lineno}, {x.name}", x.line] for x in traceback.extract_tb(tb)] - - -def format_exception(e, tb): - return {"exception": str(e), "traceback": format_traceback(tb)} - - -def get_exceptions(): - try: - return list(reversed(errors.exception_records)) - except Exception as e: - return str(e) - - def get_environment(): return {k: os.environ[k] for k in sorted(os.environ) if k in environment_whitelist} From 5cedc8f9b2b51f392e7c8f5e29286466e3bee8d6 Mon Sep 17 00:00:00 2001 From: Jabasukuriputo Wang Date: Fri, 24 Nov 2023 11:30:30 -0600 Subject: [PATCH 077/139] remove traceback in sysinfo --- modules/sysinfo.py | 1 - 1 file changed, 1 deletion(-) diff --git a/modules/sysinfo.py b/modules/sysinfo.py index 226b204d9..1d058950a 100644 --- a/modules/sysinfo.py +++ b/modules/sysinfo.py @@ -1,7 +1,6 @@ import json import os import sys -import traceback import platform import hashlib From 3a9bf4ac10d99feb81b0e637417a108d3fa5ac06 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Sun, 26 Nov 2023 08:29:12 +0300 Subject: [PATCH 078/139] move file --- {modules => extensions-builtin/hypertile}/hypertile.py | 0 1 file changed, 0 insertions(+), 0 deletions(-) rename {modules => extensions-builtin/hypertile}/hypertile.py (100%) diff --git a/modules/hypertile.py b/extensions-builtin/hypertile/hypertile.py similarity index 100% rename from modules/hypertile.py rename to extensions-builtin/hypertile/hypertile.py From d2e0c1ca132f4f0d98b77397a9f353d4ad8e7c4b Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Sun, 26 Nov 2023 10:51:45 +0300 Subject: [PATCH 079/139] rework hypertile into a built-in extension --- README.md | 1 + extensions-builtin/hypertile/hypertile.py | 215 ++++++++---------- .../hypertile/scripts/hypertile_script.py | 73 ++++++ modules/processing.py | 37 ++- modules/shared_options.py | 8 - 5 files changed, 183 insertions(+), 151 deletions(-) create mode 100644 extensions-builtin/hypertile/scripts/hypertile_script.py diff --git a/README.md b/README.md index 25ba070e7..3b3f93adc 100644 --- a/README.md +++ b/README.md @@ -174,5 +174,6 @@ Licenses for borrowed code can be found in `Settings -> Licenses` screen, and al - TAESD - Ollin Boer Bohan - https://github.com/madebyollin/taesd - LyCORIS - KohakuBlueleaf - Restart sampling - lambertae - https://github.com/Newbeeer/diffusion_restart_sampling +- Hypertile - tfernd - https://github.com/tfernd/HyperTile - Initial Gradio script - posted on 4chan by an Anonymous user. Thank you Anonymous user. - (You) diff --git a/extensions-builtin/hypertile/hypertile.py b/extensions-builtin/hypertile/hypertile.py index be898fce4..a40c13118 100644 --- a/extensions-builtin/hypertile/hypertile.py +++ b/extensions-builtin/hypertile/hypertile.py @@ -1,10 +1,13 @@ """ Hypertile module for splitting attention layers in SD-1.5 U-Net and SD-1.5 VAE -Warn : The patch works well only if the input image has a width and height that are multiples of 128 -Author : @tfernd Github : https://github.com/tfernd/HyperTile +Warn: The patch works well only if the input image has a width and height that are multiples of 128 +Original author: @tfernd Github: https://github.com/tfernd/HyperTile """ from __future__ import annotations + +import functools +from dataclasses import dataclass from typing import Callable from typing_extensions import Literal @@ -18,6 +21,19 @@ import random from einops import rearrange + +@dataclass +class HypertileParams: + depth = 0 + layer_name = "" + tile_size: int = 0 + swap_size: int = 0 + aspect_ratio: float = 1.0 + forward = None + enabled = False + + + # TODO add SD-XL layers DEPTH_LAYERS = { 0: [ @@ -176,6 +192,7 @@ DEPTH_LAYERS_XL = { RNG_INSTANCE = random.Random() + def random_divisor(value: int, min_value: int, /, max_options: int = 1) -> int: """ Returns a random divisor of value that @@ -193,10 +210,13 @@ def random_divisor(value: int, min_value: int, /, max_options: int = 1) -> int: return ns[idx] + def set_hypertile_seed(seed: int) -> None: RNG_INSTANCE.seed(seed) -def largest_tile_size_available(width:int, height:int) -> int: + +@functools.cache +def largest_tile_size_available(width: int, height: int) -> int: """ Calculates the largest tile size available for a given width and height Tile size is always a power of 2 @@ -207,6 +227,7 @@ def largest_tile_size_available(width:int, height:int) -> int: largest_tile_size_available *= 2 return largest_tile_size_available + def iterative_closest_divisors(hw:int, aspect_ratio:float) -> tuple[int, int]: """ Finds h and w such that h*w = hw and h/w = aspect_ratio @@ -219,6 +240,7 @@ def iterative_closest_divisors(hw:int, aspect_ratio:float) -> tuple[int, int]: closest_pair = pairs[ratios.index(closest_ratio)] # closest pair of divisors to aspect_ratio return closest_pair + @cache def find_hw_candidates(hw:int, aspect_ratio:float) -> tuple[int, int]: """ @@ -240,132 +262,87 @@ def find_hw_candidates(hw:int, aspect_ratio:float) -> tuple[int, int]: w = int(w_candidate) return h, w -@contextmanager -def split_attention( - layer: nn.Module, - /, - aspect_ratio: float, # width/height - tile_size: int = 128, # 128 for VAE - swap_size: int = 1, # 1 for VAE - *, - disable: bool = False, - max_depth: Literal[0, 1, 2, 3] = 0, # ! Try 0 or 1 - scale_depth: bool = True, # scale the tile-size depending on the depth - is_sdxl: bool = False, # is the model SD-XL -): - # Hijacks AttnBlock from ldm and Attention from diffusers - if disable: - logging.info(f"Attention for {layer.__class__.__qualname__} not splitted") - yield - return +def self_attn_forward(params: HypertileParams, scale_depth=True) -> Callable: - latent_tile_size = max(128, tile_size) // 8 + @wraps(params.forward) + def wrapper(*args, **kwargs): + if not params.enabled: + return params.forward(*args, **kwargs) - def self_attn_forward(forward: Callable, depth: int, layer_name: str, module: nn.Module) -> Callable: - @wraps(forward) - def wrapper(*args, **kwargs): - x = args[0] + latent_tile_size = max(128, params.tile_size) // 8 + x = args[0] - # VAE - if x.ndim == 4: - b, c, h, w = x.shape + # VAE + if x.ndim == 4: + b, c, h, w = x.shape - nh = random_divisor(h, latent_tile_size, swap_size) - nw = random_divisor(w, latent_tile_size, swap_size) + nh = random_divisor(h, latent_tile_size, params.swap_size) + nw = random_divisor(w, latent_tile_size, params.swap_size) - if nh * nw > 1: - x = rearrange(x, "b c (nh h) (nw w) -> (b nh nw) c h w", nh=nh, nw=nw) # split into nh * nw tiles + if nh * nw > 1: + x = rearrange(x, "b c (nh h) (nw w) -> (b nh nw) c h w", nh=nh, nw=nw) # split into nh * nw tiles - out = forward(x, *args[1:], **kwargs) + out = params.forward(x, *args[1:], **kwargs) - if nh * nw > 1: - out = rearrange(out, "(b nh nw) c h w -> b c (nh h) (nw w)", nh=nh, nw=nw) + if nh * nw > 1: + out = rearrange(out, "(b nh nw) c h w -> b c (nh h) (nw w)", nh=nh, nw=nw) - # U-Net - else: - hw: int = x.size(1) - h, w = find_hw_candidates(hw, aspect_ratio) - assert h * w == hw, f"Invalid aspect ratio {aspect_ratio} for input of shape {x.shape}, hw={hw}, h={h}, w={w}" - - factor = 2**depth if scale_depth else 1 - nh = random_divisor(h, latent_tile_size * factor, swap_size) - nw = random_divisor(w, latent_tile_size * factor, swap_size) - - module._split_sizes_hypertile.append((nh, nw)) # type: ignore - - if nh * nw > 1: - x = rearrange(x, "b (nh h nw w) c -> (b nh nw) (h w) c", h=h // nh, w=w // nw, nh=nh, nw=nw) - - out = forward(x, *args[1:], **kwargs) - - if nh * nw > 1: - out = rearrange(out, "(b nh nw) hw c -> b nh nw hw c", nh=nh, nw=nw) - out = rearrange(out, "b nh nw (h w) c -> b (nh h nw w) c", h=h // nh, w=w // nw) - - return out - - return wrapper - - # Handle hijacking the forward method and recovering afterwards - try: - if is_sdxl: - layers = DEPTH_LAYERS_XL + # U-Net else: - layers = DEPTH_LAYERS - for depth in range(max_depth + 1): - for layer_name, module in layer.named_modules(): + hw: int = x.size(1) + h, w = find_hw_candidates(hw, params.aspect_ratio) + assert h * w == hw, f"Invalid aspect ratio {params.aspect_ratio} for input of shape {x.shape}, hw={hw}, h={h}, w={w}" + + factor = 2 ** params.depth if scale_depth else 1 + nh = random_divisor(h, latent_tile_size * factor, params.swap_size) + nw = random_divisor(w, latent_tile_size * factor, params.swap_size) + + if nh * nw > 1: + x = rearrange(x, "b (nh h nw w) c -> (b nh nw) (h w) c", h=h // nh, w=w // nw, nh=nh, nw=nw) + + out = params.forward(x, *args[1:], **kwargs) + + if nh * nw > 1: + out = rearrange(out, "(b nh nw) hw c -> b nh nw hw c", nh=nh, nw=nw) + out = rearrange(out, "b nh nw (h w) c -> b (nh h nw w) c", h=h // nh, w=w // nw) + + return out + + return wrapper + + +def hypertile_hook_model(model: nn.Module, width, height, *, enable=False, tile_size_max=128, swap_size=1, max_depth=3, is_sdxl=False): + hypertile_layers = getattr(model, "__webui_hypertile_layers", None) + if hypertile_layers is None: + if not enable: + return + + hypertile_layers = {} + layers = DEPTH_LAYERS_XL if is_sdxl else DEPTH_LAYERS + + for depth in range(4): + for layer_name, module in model.named_modules(): if any(layer_name.endswith(try_name) for try_name in layers[depth]): - # print input shape for debugging - logging.debug(f"HyperTile hijacking attention layer at depth {depth}: {layer_name}") - # hijack - module._original_forward_hypertile = module.forward - module.forward = self_attn_forward(module.forward, depth, layer_name, module) - module._split_sizes_hypertile = [] - yield - finally: - for layer_name, module in layer.named_modules(): - # remove hijack - if hasattr(module, "_original_forward_hypertile"): - if module._split_sizes_hypertile: - logging.debug(f"layer {layer_name} splitted with ({module._split_sizes_hypertile})") - # recover - module.forward = module._original_forward_hypertile - del module._original_forward_hypertile - del module._split_sizes_hypertile + params = HypertileParams() + module.__webui_hypertile_params = params + params.forward = module.forward + params.depth = depth + params.layer_name = layer_name + module.forward = self_attn_forward(params) -def hypertile_context_vae(model:nn.Module, aspect_ratio:float, tile_size:int, opts): - """ - Returns context manager for VAE - """ - enabled = opts.hypertile_split_vae_attn - swap_size = opts.hypertile_swap_size_vae - max_depth = opts.hypertile_max_depth_vae - tile_size_max = opts.hypertile_max_tile_vae - return split_attention( - model, - aspect_ratio=aspect_ratio, - tile_size=min(tile_size, tile_size_max), - swap_size=swap_size, - disable=not enabled, - max_depth=max_depth, - is_sdxl=False, - ) + hypertile_layers[layer_name] = 1 -def hypertile_context_unet(model:nn.Module, aspect_ratio:float, tile_size:int, opts, is_sdxl:bool): - """ - Returns context manager for U-Net - """ - enabled = opts.hypertile_split_unet_attn - swap_size = opts.hypertile_swap_size_unet - max_depth = opts.hypertile_max_depth_unet - tile_size_max = opts.hypertile_max_tile_unet - return split_attention( - model, - aspect_ratio=aspect_ratio, - tile_size=min(tile_size, tile_size_max), - swap_size=swap_size, - disable=not enabled, - max_depth=max_depth, - is_sdxl=is_sdxl, - ) + model.__webui_hypertile_layers = hypertile_layers + + aspect_ratio = width / height + tile_size = min(largest_tile_size_available(width, height), tile_size_max) + + for layer_name, module in model.named_modules(): + if layer_name in hypertile_layers: + params = module.__webui_hypertile_params + + params.tile_size = tile_size + params.swap_size = swap_size + params.aspect_ratio = aspect_ratio + params.enabled = enable and params.depth <= max_depth diff --git a/extensions-builtin/hypertile/scripts/hypertile_script.py b/extensions-builtin/hypertile/scripts/hypertile_script.py new file mode 100644 index 000000000..3cc29cd1f --- /dev/null +++ b/extensions-builtin/hypertile/scripts/hypertile_script.py @@ -0,0 +1,73 @@ +import hypertile +from modules import scripts, script_callbacks, shared + + +class ScriptHypertile(scripts.Script): + name = "Hypertile" + + def title(self): + return self.name + + def show(self, is_img2img): + return scripts.AlwaysVisible + + def process(self, p, *args): + hypertile.set_hypertile_seed(p.all_seeds[0]) + + configure_hypertile(p.width, p.height, enable_unet=shared.opts.hypertile_enable_unet) + + def before_hr(self, p, *args): + configure_hypertile(p.hr_upscale_to_x, p.hr_upscale_to_y, enable_unet=shared.opts.hypertile_enable_unet_secondpass or shared.opts.hypertile_enable_unet) + + +def configure_hypertile(width, height, enable_unet=True): + hypertile.hypertile_hook_model( + shared.sd_model.first_stage_model, + width, + height, + swap_size=shared.opts.hypertile_swap_size_vae, + max_depth=shared.opts.hypertile_max_depth_vae, + tile_size_max=shared.opts.hypertile_max_tile_vae, + enable=shared.opts.hypertile_enable_vae, + ) + + hypertile.hypertile_hook_model( + shared.sd_model.model, + width, + height, + swap_size=shared.opts.hypertile_swap_size_unet, + max_depth=shared.opts.hypertile_max_depth_unet, + tile_size_max=shared.opts.hypertile_max_tile_unet, + enable=enable_unet, + is_sdxl=shared.sd_model.is_sdxl + ) + + +def on_ui_settings(): + import gradio as gr + + options = { + "hypertile_explanation": shared.OptionHTML(""" + Hypertile optimizes the self-attention layer within U-Net and VAE models, + resulting in a reduction in computation time ranging from 1 to 4 times. The larger the generated image is, the greater the + benefit. + """), + + "hypertile_enable_unet": shared.OptionInfo(False, "Enable Hypertile U-Net").info("noticeable change in details of the generated picture; if enabled, overrides the setting below"), + "hypertile_enable_unet_secondpass": shared.OptionInfo(False, "Enable Hypertile U-Net for hires fix second pass"), + "hypertile_max_depth_unet": shared.OptionInfo(3, "Hypertile U-Net max depth", gr.Slider, {"minimum": 0, "maximum": 3, "step": 1}), + "hypertile_max_tile_unet": shared.OptionInfo(256, "Hypertile U-net max tile size", gr.Slider, {"minimum": 0, "maximum": 512, "step": 16}), + "hypertile_swap_size_unet": shared.OptionInfo(3, "Hypertile U-net swap size", gr.Slider, {"minimum": 0, "maximum": 6, "step": 1}), + + "hypertile_enable_vae": shared.OptionInfo(False, "Enable Hypertile VAE").info("minimal change in the generated picture"), + "hypertile_max_depth_vae": shared.OptionInfo(3, "Hypertile VAE max depth", gr.Slider, {"minimum": 0, "maximum": 3, "step": 1}), + "hypertile_max_tile_vae": shared.OptionInfo(128, "Hypertile VAE max tile size", gr.Slider, {"minimum": 0, "maximum": 512, "step": 16}), + "hypertile_swap_size_vae": shared.OptionInfo(3, "Hypertile VAE swap size ", gr.Slider, {"minimum": 0, "maximum": 6, "step": 1}), + } + + for name, opt in options.items(): + opt.section = ('hypertile', "Hypertile") + shared.opts.add_option(name, opt) + + +script_callbacks.on_ui_settings(on_ui_settings) diff --git a/modules/processing.py b/modules/processing.py index 36c2be5e5..ac58ef869 100644 --- a/modules/processing.py +++ b/modules/processing.py @@ -24,7 +24,6 @@ from modules.shared import opts, cmd_opts, state import modules.shared as shared import modules.paths as paths import modules.face_restoration -from modules.hypertile import set_hypertile_seed, largest_tile_size_available, hypertile_context_unet, hypertile_context_vae import modules.images as images import modules.styles import modules.sd_models as sd_models @@ -861,8 +860,7 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed: p.comment(comment) p.extra_generation_params.update(model_hijack.extra_generation_params) - set_hypertile_seed(p.seed) - # add batch size + hypertile status to information to reproduce the run + if p.n_iter > 1: shared.state.job = f"Batch {n+1} out of {p.n_iter}" @@ -874,8 +872,7 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed: else: if opts.sd_vae_decode_method != 'Full': p.extra_generation_params['VAE Decoder'] = opts.sd_vae_decode_method - with hypertile_context_vae(p.sd_model.first_stage_model, aspect_ratio=p.width / p.height, tile_size=largest_tile_size_available(p.width, p.height), opts=shared.opts): - x_samples_ddim = decode_latent_batch(p.sd_model, samples_ddim, target_device=devices.cpu, check_for_nans=True) + x_samples_ddim = decode_latent_batch(p.sd_model, samples_ddim, target_device=devices.cpu, check_for_nans=True) x_samples_ddim = torch.stack(x_samples_ddim).float() x_samples_ddim = torch.clamp((x_samples_ddim + 1.0) / 2.0, min=0.0, max=1.0) @@ -1141,25 +1138,23 @@ class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing): def sample(self, conditioning, unconditional_conditioning, seeds, subseeds, subseed_strength, prompts): self.sampler = sd_samplers.create_sampler(self.sampler_name, self.sd_model) - aspect_ratio = self.width / self.height + x = self.rng.next() - tile_size = largest_tile_size_available(self.width, self.height) - with hypertile_context_vae(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=tile_size, opts=shared.opts): - with hypertile_context_unet(self.sd_model.model, aspect_ratio=aspect_ratio, tile_size=tile_size, is_sdxl=shared.sd_model.is_sdxl, opts=shared.opts): - samples = self.sampler.sample(self, x, conditioning, unconditional_conditioning, image_conditioning=self.txt2img_image_conditioning(x)) + samples = self.sampler.sample(self, x, conditioning, unconditional_conditioning, image_conditioning=self.txt2img_image_conditioning(x)) del x + if not self.enable_hr: return samples devices.torch_gc() if self.latent_scale_mode is None: - with hypertile_context_vae(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=tile_size, opts=shared.opts): - decoded_samples = torch.stack(decode_latent_batch(self.sd_model, samples, target_device=devices.cpu, check_for_nans=True)).to(dtype=torch.float32) + decoded_samples = torch.stack(decode_latent_batch(self.sd_model, samples, target_device=devices.cpu, check_for_nans=True)).to(dtype=torch.float32) else: decoded_samples = None with sd_models.SkipWritingToConfig(): sd_models.reload_model_weights(info=self.hr_checkpoint_info) + return self.sample_hr_pass(samples, decoded_samples, seeds, subseeds, subseed_strength, prompts) def sample_hr_pass(self, samples, decoded_samples, seeds, subseeds, subseed_strength, prompts): @@ -1244,18 +1239,15 @@ class StableDiffusionProcessingTxt2Img(StableDiffusionProcessing): if self.scripts is not None: self.scripts.before_hr(self) - tile_size = largest_tile_size_available(target_width, target_height) - aspect_ratio = self.width / self.height - with hypertile_context_vae(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=tile_size, opts=shared.opts): - with hypertile_context_unet(self.sd_model.model, aspect_ratio=aspect_ratio, tile_size=tile_size, is_sdxl=shared.sd_model.is_sdxl, opts=shared.opts): - samples = self.sampler.sample_img2img(self, samples, noise, self.hr_c, self.hr_uc, steps=self.hr_second_pass_steps or self.steps, image_conditioning=image_conditioning) + + samples = self.sampler.sample_img2img(self, samples, noise, self.hr_c, self.hr_uc, steps=self.hr_second_pass_steps or self.steps, image_conditioning=image_conditioning) sd_models.apply_token_merging(self.sd_model, self.get_token_merging_ratio()) self.sampler = None devices.torch_gc() - with hypertile_context_vae(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=tile_size, opts=shared.opts): - decoded_samples = decode_latent_batch(self.sd_model, samples, target_device=devices.cpu, check_for_nans=True) + + decoded_samples = decode_latent_batch(self.sd_model, samples, target_device=devices.cpu, check_for_nans=True) self.is_hr_pass = False return decoded_samples @@ -1532,11 +1524,8 @@ class StableDiffusionProcessingImg2Img(StableDiffusionProcessing): if self.initial_noise_multiplier != 1.0: self.extra_generation_params["Noise multiplier"] = self.initial_noise_multiplier x *= self.initial_noise_multiplier - aspect_ratio = self.width / self.height - tile_size = largest_tile_size_available(self.width, self.height) - with hypertile_context_vae(self.sd_model.first_stage_model, aspect_ratio=aspect_ratio, tile_size=tile_size, opts=shared.opts): - with hypertile_context_unet(self.sd_model.model, aspect_ratio=aspect_ratio, tile_size=tile_size, is_sdxl=shared.sd_model.is_sdxl, opts=shared.opts): - samples = self.sampler.sample_img2img(self, self.init_latent, x, conditioning, unconditional_conditioning, image_conditioning=self.image_conditioning) + + samples = self.sampler.sample_img2img(self, self.init_latent, x, conditioning, unconditional_conditioning, image_conditioning=self.image_conditioning) if self.mask is not None: samples = samples * self.nmask + self.init_latent * self.mask diff --git a/modules/shared_options.py b/modules/shared_options.py index 28a489069..d40db5306 100644 --- a/modules/shared_options.py +++ b/modules/shared_options.py @@ -200,14 +200,6 @@ options_templates.update(options_section(('optimizations', "Optimizations"), { "pad_cond_uncond": OptionInfo(False, "Pad prompt/negative prompt to be same length", infotext='Pad conds').info("improves performance when prompt and negative prompt have different lengths; changes seeds"), "persistent_cond_cache": OptionInfo(True, "Persistent cond cache").info("do not recalculate conds from prompts if prompts have not changed since previous calculation"), "batch_cond_uncond": OptionInfo(True, "Batch cond/uncond").info("do both conditional and unconditional denoising in one batch; uses a bit more VRAM during sampling, but improves speed; previously this was controlled by --always-batch-cond-uncond comandline argument"), - "hypertile_split_unet_attn" : OptionInfo(False, "Split attention in Unet with HyperTile").link("Github", "https://github.com/tfernd/HyperTile").info("improves performance; changes behavior, but deterministic"), - "hypertile_split_vae_attn": OptionInfo(False, "Split attention in VAE with HyperTile").link("Github", "https://github.com/tfernd/HyperTile").info("improves performance; changes behavior, but deterministic"), - "hypertile_max_depth_vae" : OptionInfo(3, "Max depth for VAE HyperTile hijack", gr.Slider, {"minimum": 0, "maximum": 3, "step": 1}).link("Github", "https://github.com/tfernd/HyperTile"), - "hypertile_max_depth_unet" : OptionInfo(3, "Max depth for Unet HyperTile hijack", gr.Slider, {"minimum": 0, "maximum": 3, "step": 1}).link("Github", "https://github.com/tfernd/HyperTile"), - "hypertile_max_tile_vae" : OptionInfo(128, "Max tile size for VAE HyperTile hijack", gr.Slider, {"minimum": 0, "maximum": 512, "step": 16}).link("Github", "https://github.com/tfernd/HyperTile"), - "hypertile_max_tile_unet" : OptionInfo(256, "Max tile size for Unet HyperTile hijack", gr.Slider, {"minimum": 0, "maximum": 512, "step": 16}).link("Github", "https://github.com/tfernd/HyperTile"), - "hypertile_swap_size_unet": OptionInfo(3, "Swap size for Unet HyperTile hijack", gr.Slider, {"minimum": 0, "maximum": 6, "step": 1}).link("Github", "https://github.com/tfernd/HyperTile"), - "hypertile_swap_size_vae": OptionInfo(3, "Swap size for VAE HyperTile hijack", gr.Slider, {"minimum": 0, "maximum": 6, "step": 1}).link("Github", "https://github.com/tfernd/HyperTile"), })) options_templates.update(options_section(('compatibility', "Compatibility"), { From c5a0c59a83c950c64bc44427d3478aaa78c296cf Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Sun, 26 Nov 2023 11:36:17 +0300 Subject: [PATCH 080/139] do not save HTML explanations from options page to config --- modules/options.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/modules/options.py b/modules/options.py index ab40aff73..7703d80ec 100644 --- a/modules/options.py +++ b/modules/options.py @@ -76,7 +76,7 @@ class Options: def __init__(self, data_labels: dict[str, OptionInfo], restricted_opts): self.data_labels = data_labels - self.data = {k: v.default for k, v in self.data_labels.items()} + self.data = {k: v.default for k, v in self.data_labels.items() if not v.do_not_save} self.restricted_opts = restricted_opts def __setattr__(self, key, value): @@ -210,7 +210,7 @@ class Options: def add_option(self, key, info): self.data_labels[key] = info - if key not in self.data: + if key not in self.data and not info.do_not_save: self.data[key] = info.default def reorder(self): From d1750e5eca6fd95db3516928cad18b32e557f56f Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Sun, 26 Nov 2023 11:37:12 +0300 Subject: [PATCH 081/139] fix linter errors --- extensions-builtin/hypertile/hypertile.py | 3 --- 1 file changed, 3 deletions(-) diff --git a/extensions-builtin/hypertile/hypertile.py b/extensions-builtin/hypertile/hypertile.py index a40c13118..feb02fd27 100644 --- a/extensions-builtin/hypertile/hypertile.py +++ b/extensions-builtin/hypertile/hypertile.py @@ -9,11 +9,8 @@ from __future__ import annotations import functools from dataclasses import dataclass from typing import Callable -from typing_extensions import Literal -import logging from functools import wraps, cache -from contextlib import contextmanager import math import torch.nn as nn From 2a40d3c603448d15e209814366f2d6ab25e52398 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Sun, 26 Nov 2023 14:58:47 +0300 Subject: [PATCH 082/139] compact prompt layout: preserve scroll when switching between lora tabs --- javascript/extraNetworks.js | 4 ++++ modules/ui_extra_networks.py | 5 ++++- style.css | 12 ++++++++++-- 3 files changed, 18 insertions(+), 3 deletions(-) diff --git a/javascript/extraNetworks.js b/javascript/extraNetworks.js index a1bf29a8c..a787372cf 100644 --- a/javascript/extraNetworks.js +++ b/javascript/extraNetworks.js @@ -130,6 +130,10 @@ function extraNetworksMovePromptToTab(tabname, id, showPrompt, showNegativePromp } else { promptContainer.insertBefore(prompt, promptContainer.firstChild); } + + if (elem) { + elem.classList.toggle('extra-page-prompts-active', showNegativePrompt || showPrompt); + } } diff --git a/modules/ui_extra_networks.py b/modules/ui_extra_networks.py index f03e20337..f3b23cc9c 100644 --- a/modules/ui_extra_networks.py +++ b/modules/ui_extra_networks.py @@ -370,6 +370,9 @@ def create_ui(interface: gr.Blocks, unrelated_tabs, tabname): for page in ui.stored_extra_pages: with gr.Tab(page.title, elem_id=f"{tabname}_{page.id_page}", elem_classes=["extra-page"]) as tab: + with gr.Column(elem_id=f"{tabname}_{page.id_page}_prompts", elem_classes=["extra-page-prompts"]): + pass + elem_id = f"{tabname}_{page.id_page}_cards_html" page_elem = gr.HTML('Loading...', elem_id=elem_id) ui.pages.append(page_elem) @@ -400,7 +403,7 @@ def create_ui(interface: gr.Blocks, unrelated_tabs, tabname): allow_prompt = "true" if page.allow_prompt else "false" allow_negative_prompt = "true" if page.allow_negative_prompt else "false" - jscode = 'extraNetworksTabSelected("' + tabname + '", "' + f"{tabname}_{page.id_page}" + '", ' + allow_prompt + ', ' + allow_negative_prompt + ');' + jscode = 'extraNetworksTabSelected("' + tabname + '", "' + f"{tabname}_{page.id_page}_prompts" + '", ' + allow_prompt + ', ' + allow_negative_prompt + ');' tab.select(fn=lambda: [gr.update(visible=True) for _ in tab_controls], _js='function(){ ' + jscode + ' }', inputs=[], outputs=tab_controls, show_progress=False) diff --git a/style.css b/style.css index 731620226..f8b42636d 100644 --- a/style.css +++ b/style.css @@ -840,8 +840,16 @@ footer { /* extra networks UI */ -.extra-page .prompt{ - margin: 0 0 0.5em 0; +.extra-page > div.gap{ + gap: 0; +} + +.extra-page-prompts{ + margin-bottom: 0; +} + +.extra-page-prompts.extra-page-prompts-active{ + margin-bottom: 1em; } .extra-network-cards{ From a15dd151ffb4d11556028b34561058bc44930427 Mon Sep 17 00:00:00 2001 From: w-e-w <40751091+w-e-w@users.noreply.github.com> Date: Sun, 26 Nov 2023 21:55:50 +0900 Subject: [PATCH 083/139] json.dump(ensure_ascii=False) improve json readability --- modules/cache.py | 2 +- modules/options.py | 2 +- modules/ui_extensions.py | 2 +- modules/ui_extra_networks_user_metadata.py | 2 +- modules/ui_loadsave.py | 2 +- 5 files changed, 5 insertions(+), 5 deletions(-) diff --git a/modules/cache.py b/modules/cache.py index ff26a2132..2d37e7b99 100644 --- a/modules/cache.py +++ b/modules/cache.py @@ -32,7 +32,7 @@ def dump_cache(): with cache_lock: cache_filename_tmp = cache_filename + "-" with open(cache_filename_tmp, "w", encoding="utf8") as file: - json.dump(cache_data, file, indent=4) + json.dump(cache_data, file, indent=4, ensure_ascii=False) os.replace(cache_filename_tmp, cache_filename) diff --git a/modules/options.py b/modules/options.py index 7703d80ec..40cb47991 100644 --- a/modules/options.py +++ b/modules/options.py @@ -158,7 +158,7 @@ class Options: assert not cmd_opts.freeze_settings, "saving settings is disabled" with open(filename, "w", encoding="utf8") as file: - json.dump(self.data, file, indent=4) + json.dump(self.data, file, indent=4, ensure_ascii=False) def same_type(self, x, y): if x is None or y is None: diff --git a/modules/ui_extensions.py b/modules/ui_extensions.py index c0a73b573..96dc9db2c 100644 --- a/modules/ui_extensions.py +++ b/modules/ui_extensions.py @@ -65,7 +65,7 @@ def save_config_state(name): filename = os.path.join(config_states_dir, f"{timestamp}_{name}.json") print(f"Saving backup of webui/extension state to {filename}.") with open(filename, "w", encoding="utf-8") as f: - json.dump(current_config_state, f, indent=4) + json.dump(current_config_state, f, indent=4, ensure_ascii=False) config_states.list_config_states() new_value = next(iter(config_states.all_config_states.keys()), "Current") new_choices = ["Current"] + list(config_states.all_config_states.keys()) diff --git a/modules/ui_extra_networks_user_metadata.py b/modules/ui_extra_networks_user_metadata.py index bfec140cc..36a807fcd 100644 --- a/modules/ui_extra_networks_user_metadata.py +++ b/modules/ui_extra_networks_user_metadata.py @@ -134,7 +134,7 @@ class UserMetadataEditor: basename, ext = os.path.splitext(filename) with open(basename + '.json', "w", encoding="utf8") as file: - json.dump(metadata, file, indent=4) + json.dump(metadata, file, indent=4, ensure_ascii=False) def save_user_metadata(self, name, desc, notes): user_metadata = self.get_user_metadata(name) diff --git a/modules/ui_loadsave.py b/modules/ui_loadsave.py index eb20ff258..7826786cc 100644 --- a/modules/ui_loadsave.py +++ b/modules/ui_loadsave.py @@ -141,7 +141,7 @@ class UiLoadsave: def write_to_file(self, current_ui_settings): with open(self.filename, "w", encoding="utf8") as file: - json.dump(current_ui_settings, file, indent=4) + json.dump(current_ui_settings, file, indent=4, ensure_ascii=False) def dump_defaults(self): """saves default values to a file unless tjhe file is present and there was an error loading default values at start""" From f0f100e67b78f686dc73cf3c8cad422e45cc9b8a Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Sun, 26 Nov 2023 17:56:16 +0300 Subject: [PATCH 084/139] add categories to settings --- javascript/settings.js | 25 +++++++++++++ modules/options.py | 75 +++++++++++++++++++++++++++++++++++---- modules/shared_options.py | 49 ++++++++++++++----------- style.css | 9 +++++ 4 files changed, 130 insertions(+), 28 deletions(-) diff --git a/javascript/settings.js b/javascript/settings.js index 4e79ec003..e6009290a 100644 --- a/javascript/settings.js +++ b/javascript/settings.js @@ -44,3 +44,28 @@ onUiLoaded(function() { buttonShowAllPages.addEventListener("click", settingsShowAllTabs); }); + + +onOptionsChanged(function() { + if (gradioApp().querySelector('#settings .settings-category')) return; + + var sectionMap = {}; + gradioApp().querySelectorAll('#settings > div > button').forEach(function(x) { + sectionMap[x.textContent.trim()] = x; + }); + + opts._categories.forEach(function(x) { + var section = x[0]; + var category = x[1]; + + var span = document.createElement('SPAN'); + span.textContent = category; + span.className = 'settings-category'; + + var sectionElem = sectionMap[section]; + if (!sectionElem) return; + + sectionElem.parentElement.insertBefore(span, sectionElem); + }); +}); + diff --git a/modules/options.py b/modules/options.py index 40cb47991..4fead690c 100644 --- a/modules/options.py +++ b/modules/options.py @@ -1,5 +1,6 @@ import json import sys +from dataclasses import dataclass import gradio as gr @@ -8,13 +9,14 @@ from modules.shared_cmd_options import cmd_opts class OptionInfo: - def __init__(self, default=None, label="", component=None, component_args=None, onchange=None, section=None, refresh=None, comment_before='', comment_after='', infotext=None, restrict_api=False): + def __init__(self, default=None, label="", component=None, component_args=None, onchange=None, section=None, refresh=None, comment_before='', comment_after='', infotext=None, restrict_api=False, category_id=None): self.default = default self.label = label self.component = component self.component_args = component_args self.onchange = onchange self.section = section + self.category_id = category_id self.refresh = refresh self.do_not_save = False @@ -63,7 +65,11 @@ class OptionHTML(OptionInfo): def options_section(section_identifier, options_dict): for v in options_dict.values(): - v.section = section_identifier + if len(section_identifier) == 2: + v.section = section_identifier + elif len(section_identifier) == 3: + v.section = section_identifier[0:2] + v.category_id = section_identifier[2] return options_dict @@ -206,6 +212,17 @@ class Options: d = {k: self.data.get(k, v.default) for k, v in self.data_labels.items()} d["_comments_before"] = {k: v.comment_before for k, v in self.data_labels.items() if v.comment_before is not None} d["_comments_after"] = {k: v.comment_after for k, v in self.data_labels.items() if v.comment_after is not None} + + item_categories = {} + for item in self.data_labels.values(): + category = categories.mapping.get(item.category_id) + category = "Uncategorized" if category is None else category.label + if category not in item_categories: + item_categories[category] = item.section[1] + + # _categories is a list of pairs: [section, category]. Each section (a setting page) will get a special heading above it with the category as text. + d["_categories"] = [[v, k] for k, v in item_categories.items()] + [["Defaults", "Other"]] + return json.dumps(d) def add_option(self, key, info): @@ -214,15 +231,40 @@ class Options: self.data[key] = info.default def reorder(self): - """reorder settings so that all items related to section always go together""" + """Reorder settings so that: + - all items related to section always go together + - all sections belonging to a category go together + - sections inside a category are ordered alphabetically + - categories are ordered by creation order + + Category is a superset of sections: for category "postprocessing" there could be multiple sections: "face restoration", "upscaling". + + This function also changes items' category_id so that all items belonging to a section have the same category_id. + """ + + category_ids = {} + section_categories = {} - section_ids = {} settings_items = self.data_labels.items() for _, item in settings_items: - if item.section not in section_ids: - section_ids[item.section] = len(section_ids) + if item.section not in section_categories: + section_categories[item.section] = item.category_id - self.data_labels = dict(sorted(settings_items, key=lambda x: section_ids[x[1].section])) + for _, item in settings_items: + item.category_id = section_categories.get(item.section) + + for category_id in categories.mapping: + if category_id not in category_ids: + category_ids[category_id] = len(category_ids) + + def sort_key(x): + item: OptionInfo = x[1] + category_order = category_ids.get(item.category_id, len(category_ids)) + section_order = item.section[1] + + return category_order, section_order + + self.data_labels = dict(sorted(settings_items, key=sort_key)) def cast_value(self, key, value): """casts an arbitrary to the same type as this setting's value with key @@ -245,3 +287,22 @@ class Options: value = expected_type(value) return value + + +@dataclass +class OptionsCategory: + id: str + label: str + +class OptionsCategories: + def __init__(self): + self.mapping = {} + + def register_category(self, category_id, label): + if category_id in self.mapping: + return category_id + + self.mapping[category_id] = OptionsCategory(category_id, label) + + +categories = OptionsCategories() diff --git a/modules/shared_options.py b/modules/shared_options.py index 9bcd7914b..04e68a712 100644 --- a/modules/shared_options.py +++ b/modules/shared_options.py @@ -3,7 +3,7 @@ import gradio as gr from modules import localization, ui_components, shared_items, shared, interrogate, shared_gradio_themes from modules.paths_internal import models_path, script_path, data_path, sd_configs_path, sd_default_config, sd_model_file, default_sd_model_file, extensions_dir, extensions_builtin_dir # noqa: F401 from modules.shared_cmd_options import cmd_opts -from modules.options import options_section, OptionInfo, OptionHTML +from modules.options import options_section, OptionInfo, OptionHTML, categories options_templates = {} hide_dirs = shared.hide_dirs @@ -21,7 +21,14 @@ restricted_opts = { "outdir_init_images" } -options_templates.update(options_section(('saving-images', "Saving images/grids"), { +categories.register_category("saving", "Saving images") +categories.register_category("sd", "Stable Diffusion") +categories.register_category("ui", "User Interface") +categories.register_category("system", "System") +categories.register_category("postprocessing", "Postprocessing") +categories.register_category("training", "Training") + +options_templates.update(options_section(('saving-images', "Saving images/grids", "saving"), { "samples_save": OptionInfo(True, "Always save all generated images"), "samples_format": OptionInfo('png', 'File format for images'), "samples_filename_pattern": OptionInfo("", "Images filename pattern", component_args=hide_dirs).link("wiki", "https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Custom-Images-Filename-Name-and-Subdirectory"), @@ -67,7 +74,7 @@ options_templates.update(options_section(('saving-images', "Saving images/grids" "notification_volume": OptionInfo(100, "Notification sound volume", gr.Slider, {"minimum": 0, "maximum": 100, "step": 1}).info("in %"), })) -options_templates.update(options_section(('saving-paths', "Paths for saving"), { +options_templates.update(options_section(('saving-paths', "Paths for saving", "saving"), { "outdir_samples": OptionInfo("", "Output directory for images; if empty, defaults to three directories below", component_args=hide_dirs), "outdir_txt2img_samples": OptionInfo("outputs/txt2img-images", 'Output directory for txt2img images', component_args=hide_dirs), "outdir_img2img_samples": OptionInfo("outputs/img2img-images", 'Output directory for img2img images', component_args=hide_dirs), @@ -79,7 +86,7 @@ options_templates.update(options_section(('saving-paths', "Paths for saving"), { "outdir_init_images": OptionInfo("outputs/init-images", "Directory for saving init images when using img2img", component_args=hide_dirs), })) -options_templates.update(options_section(('saving-to-dirs', "Saving to a directory"), { +options_templates.update(options_section(('saving-to-dirs', "Saving to a directory", "saving"), { "save_to_dirs": OptionInfo(True, "Save images to a subdirectory"), "grid_save_to_dirs": OptionInfo(True, "Save grids to a subdirectory"), "use_save_to_dirs_for_ui": OptionInfo(False, "When using \"Save\" button, save images to a subdirectory"), @@ -87,21 +94,21 @@ options_templates.update(options_section(('saving-to-dirs', "Saving to a directo "directories_max_prompt_words": OptionInfo(8, "Max prompt words for [prompt_words] pattern", gr.Slider, {"minimum": 1, "maximum": 20, "step": 1, **hide_dirs}), })) -options_templates.update(options_section(('upscaling', "Upscaling"), { +options_templates.update(options_section(('upscaling', "Upscaling", "postprocessing"), { "ESRGAN_tile": OptionInfo(192, "Tile size for ESRGAN upscalers.", gr.Slider, {"minimum": 0, "maximum": 512, "step": 16}).info("0 = no tiling"), "ESRGAN_tile_overlap": OptionInfo(8, "Tile overlap for ESRGAN upscalers.", gr.Slider, {"minimum": 0, "maximum": 48, "step": 1}).info("Low values = visible seam"), "realesrgan_enabled_models": OptionInfo(["R-ESRGAN 4x+", "R-ESRGAN 4x+ Anime6B"], "Select which Real-ESRGAN models to show in the web UI.", gr.CheckboxGroup, lambda: {"choices": shared_items.realesrgan_models_names()}), "upscaler_for_img2img": OptionInfo(None, "Upscaler for img2img", gr.Dropdown, lambda: {"choices": [x.name for x in shared.sd_upscalers]}), })) -options_templates.update(options_section(('face-restoration', "Face restoration"), { +options_templates.update(options_section(('face-restoration', "Face restoration", "postprocessing"), { "face_restoration": OptionInfo(False, "Restore faces", infotext='Face restoration').info("will use a third-party model on generation result to reconstruct faces"), "face_restoration_model": OptionInfo("CodeFormer", "Face restoration model", gr.Radio, lambda: {"choices": [x.name() for x in shared.face_restorers]}), "code_former_weight": OptionInfo(0.5, "CodeFormer weight", gr.Slider, {"minimum": 0, "maximum": 1, "step": 0.01}).info("0 = maximum effect; 1 = minimum effect"), "face_restoration_unload": OptionInfo(False, "Move face restoration model from VRAM into RAM after processing"), })) -options_templates.update(options_section(('system', "System"), { +options_templates.update(options_section(('system', "System", "system"), { "auto_launch_browser": OptionInfo("Local", "Automatically open webui in browser on startup", gr.Radio, lambda: {"choices": ["Disable", "Local", "Remote"]}), "enable_console_prompts": OptionInfo(shared.cmd_opts.enable_console_prompts, "Print prompts to console when generating with txt2img and img2img."), "show_warnings": OptionInfo(False, "Show warnings in console.").needs_reload_ui(), @@ -116,13 +123,13 @@ options_templates.update(options_section(('system', "System"), { "dump_stacks_on_signal": OptionInfo(False, "Print stack traces before exiting the program with ctrl+c."), })) -options_templates.update(options_section(('API', "API"), { +options_templates.update(options_section(('API', "API", "system"), { "api_enable_requests": OptionInfo(True, "Allow http:// and https:// URLs for input images in API", restrict_api=True), "api_forbid_local_requests": OptionInfo(True, "Forbid URLs to local resources", restrict_api=True), "api_useragent": OptionInfo("", "User agent for requests", restrict_api=True), })) -options_templates.update(options_section(('training', "Training"), { +options_templates.update(options_section(('training', "Training", "training"), { "unload_models_when_training": OptionInfo(False, "Move VAE and CLIP to RAM when training if possible. Saves VRAM."), "pin_memory": OptionInfo(False, "Turn on pin_memory for DataLoader. Makes training slightly faster but can increase memory usage."), "save_optimizer_state": OptionInfo(False, "Saves Optimizer state as separate *.optim file. Training of embedding or HN can be resumed with the matching optim file."), @@ -137,7 +144,7 @@ options_templates.update(options_section(('training', "Training"), { "training_tensorboard_flush_every": OptionInfo(120, "How often, in seconds, to flush the pending tensorboard events and summaries to disk."), })) -options_templates.update(options_section(('sd', "Stable Diffusion"), { +options_templates.update(options_section(('sd', "Stable Diffusion", "sd"), { "sd_model_checkpoint": OptionInfo(None, "Stable Diffusion checkpoint", gr.Dropdown, lambda: {"choices": shared_items.list_checkpoint_tiles(shared.opts.sd_checkpoint_dropdown_use_short)}, refresh=shared_items.refresh_checkpoints, infotext='Model hash'), "sd_checkpoints_limit": OptionInfo(1, "Maximum number of checkpoints loaded at the same time", gr.Slider, {"minimum": 1, "maximum": 10, "step": 1}), "sd_checkpoints_keep_in_cpu": OptionInfo(True, "Only keep one model on device").info("will keep models other than the currently used one in RAM rather than VRAM"), @@ -154,14 +161,14 @@ options_templates.update(options_section(('sd', "Stable Diffusion"), { "hires_fix_refiner_pass": OptionInfo("second pass", "Hires fix: which pass to enable refiner for", gr.Radio, {"choices": ["first pass", "second pass", "both passes"]}, infotext="Hires refiner"), })) -options_templates.update(options_section(('sdxl', "Stable Diffusion XL"), { +options_templates.update(options_section(('sdxl', "Stable Diffusion XL", "sd"), { "sdxl_crop_top": OptionInfo(0, "crop top coordinate"), "sdxl_crop_left": OptionInfo(0, "crop left coordinate"), "sdxl_refiner_low_aesthetic_score": OptionInfo(2.5, "SDXL low aesthetic score", gr.Number).info("used for refiner model negative prompt"), "sdxl_refiner_high_aesthetic_score": OptionInfo(6.0, "SDXL high aesthetic score", gr.Number).info("used for refiner model prompt"), })) -options_templates.update(options_section(('vae', "VAE"), { +options_templates.update(options_section(('vae', "VAE", "sd"), { "sd_vae_explanation": OptionHTML(""" VAE is a neural network that transforms a standard RGB image into latent space representation and back. Latent space representation is what stable diffusion is working on during sampling @@ -176,7 +183,7 @@ For img2img, VAE is used to process user's input image before the sampling, and "sd_vae_decode_method": OptionInfo("Full", "VAE type for decode", gr.Radio, {"choices": ["Full", "TAESD"]}, infotext='VAE Decoder').info("method to decode latent to image"), })) -options_templates.update(options_section(('img2img', "img2img"), { +options_templates.update(options_section(('img2img', "img2img", "sd"), { "inpainting_mask_weight": OptionInfo(1.0, "Inpainting conditioning mask strength", gr.Slider, {"minimum": 0.0, "maximum": 1.0, "step": 0.01}, infotext='Conditional mask weight'), "initial_noise_multiplier": OptionInfo(1.0, "Noise multiplier for img2img", gr.Slider, {"minimum": 0.0, "maximum": 1.5, "step": 0.001}, infotext='Noise multiplier'), "img2img_extra_noise": OptionInfo(0.0, "Extra noise multiplier for img2img and hires fix", gr.Slider, {"minimum": 0.0, "maximum": 1.0, "step": 0.01}, infotext='Extra noise').info("0 = disabled (default); should be lower than denoising strength"), @@ -192,7 +199,7 @@ options_templates.update(options_section(('img2img', "img2img"), { "img2img_batch_show_results_limit": OptionInfo(32, "Show the first N batch img2img results in UI", gr.Slider, {"minimum": -1, "maximum": 1000, "step": 1}).info('0: disable, -1: show all images. Too many images can cause lag'), })) -options_templates.update(options_section(('optimizations', "Optimizations"), { +options_templates.update(options_section(('optimizations', "Optimizations", "sd"), { "cross_attention_optimization": OptionInfo("Automatic", "Cross attention optimization", gr.Dropdown, lambda: {"choices": shared_items.cross_attention_optimizations()}), "s_min_uncond": OptionInfo(0.0, "Negative Guidance minimum sigma", gr.Slider, {"minimum": 0.0, "maximum": 15.0, "step": 0.01}).link("PR", "https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/9177").info("skip negative prompt for some steps when the image is almost ready; 0=disable, higher=faster"), "token_merging_ratio": OptionInfo(0.0, "Token merging ratio", gr.Slider, {"minimum": 0.0, "maximum": 0.9, "step": 0.1}, infotext='Token merging ratio').link("PR", "https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/9256").info("0=disable, higher=faster"), @@ -203,7 +210,7 @@ options_templates.update(options_section(('optimizations', "Optimizations"), { "batch_cond_uncond": OptionInfo(True, "Batch cond/uncond").info("do both conditional and unconditional denoising in one batch; uses a bit more VRAM during sampling, but improves speed; previously this was controlled by --always-batch-cond-uncond comandline argument"), })) -options_templates.update(options_section(('compatibility', "Compatibility"), { +options_templates.update(options_section(('compatibility', "Compatibility", "sd"), { "use_old_emphasis_implementation": OptionInfo(False, "Use old emphasis implementation. Can be useful to reproduce old seeds."), "use_old_karras_scheduler_sigmas": OptionInfo(False, "Use old karras scheduler sigmas (0.1 to 10)."), "no_dpmpp_sde_batch_determinism": OptionInfo(False, "Do not make DPM++ SDE deterministic across different batch sizes."), @@ -228,7 +235,7 @@ options_templates.update(options_section(('interrogate', "Interrogate"), { "deepbooru_filter_tags": OptionInfo("", "deepbooru: filter out those tags").info("separate by comma"), })) -options_templates.update(options_section(('extra_networks', "Extra Networks"), { +options_templates.update(options_section(('extra_networks', "Extra Networks", "sd"), { "extra_networks_show_hidden_directories": OptionInfo(True, "Show hidden directories").info("directory is hidden if its name starts with \".\"."), "extra_networks_hidden_models": OptionInfo("When searched", "Show cards for models in hidden directories", gr.Radio, {"choices": ["Always", "When searched", "Never"]}).info('"When searched" option will only show the item when the search string has 4 characters or more'), "extra_networks_default_multiplier": OptionInfo(1.0, "Default multiplier for extra networks", gr.Slider, {"minimum": 0.0, "maximum": 2.0, "step": 0.01}), @@ -245,7 +252,7 @@ options_templates.update(options_section(('extra_networks', "Extra Networks"), { "sd_hypernetwork": OptionInfo("None", "Add hypernetwork to prompt", gr.Dropdown, lambda: {"choices": ["None", *shared.hypernetworks]}, refresh=shared_items.reload_hypernetworks), })) -options_templates.update(options_section(('ui', "User interface"), { +options_templates.update(options_section(('ui', "User interface", "ui"), { "localization": OptionInfo("None", "Localization", gr.Dropdown, lambda: {"choices": ["None"] + list(localization.localizations.keys())}, refresh=lambda: localization.list_localizations(cmd_opts.localizations_dir)).needs_reload_ui(), "gradio_theme": OptionInfo("Default", "Gradio theme", ui_components.DropdownEditable, lambda: {"choices": ["Default"] + shared_gradio_themes.gradio_hf_hub_themes}).info("you can also manually enter any of themes from the gallery.").needs_reload_ui(), "gradio_themes_cache": OptionInfo(True, "Cache gradio themes locally").info("disable to update the selected Gradio theme"), @@ -280,7 +287,7 @@ options_templates.update(options_section(('ui', "User interface"), { })) -options_templates.update(options_section(('infotext', "Infotext"), { +options_templates.update(options_section(('infotext', "Infotext", "ui"), { "add_model_hash_to_info": OptionInfo(True, "Add model hash to generation information"), "add_model_name_to_info": OptionInfo(True, "Add model name to generation information"), "add_user_name_to_info": OptionInfo(False, "Add user name to generation information when authenticated"), @@ -295,7 +302,7 @@ options_templates.update(options_section(('infotext', "Infotext"), { })) -options_templates.update(options_section(('ui', "Live previews"), { +options_templates.update(options_section(('ui', "Live previews", "ui"), { "show_progressbar": OptionInfo(True, "Show progressbar"), "live_previews_enable": OptionInfo(True, "Show live previews of the created image"), "live_previews_image_format": OptionInfo("png", "Live preview file format", gr.Radio, {"choices": ["jpeg", "png", "webp"]}), @@ -308,7 +315,7 @@ options_templates.update(options_section(('ui', "Live previews"), { "live_preview_fast_interrupt": OptionInfo(False, "Return image with chosen live preview method on interrupt").info("makes interrupts faster"), })) -options_templates.update(options_section(('sampler-params', "Sampler parameters"), { +options_templates.update(options_section(('sampler-params', "Sampler parameters", "sd"), { "hide_samplers": OptionInfo([], "Hide samplers in user interface", gr.CheckboxGroup, lambda: {"choices": [x.name for x in shared_items.list_samplers()]}).needs_reload_ui(), "eta_ddim": OptionInfo(0.0, "Eta for DDIM", gr.Slider, {"minimum": 0.0, "maximum": 1.0, "step": 0.01}, infotext='Eta DDIM').info("noise multiplier; higher = more unpredictable results"), "eta_ancestral": OptionInfo(1.0, "Eta for k-diffusion samplers", gr.Slider, {"minimum": 0.0, "maximum": 1.0, "step": 0.01}, infotext='Eta').info("noise multiplier; currently only applies to ancestral samplers (i.e. Euler a) and SDE samplers"), @@ -330,7 +337,7 @@ options_templates.update(options_section(('sampler-params', "Sampler parameters" 'uni_pc_lower_order_final': OptionInfo(True, "UniPC lower order final", infotext='UniPC lower order final'), })) -options_templates.update(options_section(('postprocessing', "Postprocessing"), { +options_templates.update(options_section(('postprocessing', "Postprocessing", "postprocessing"), { 'postprocessing_enable_in_main_ui': OptionInfo([], "Enable postprocessing operations in txt2img and img2img tabs", ui_components.DropdownMulti, lambda: {"choices": [x.name for x in shared_items.postprocessing_scripts()]}), 'postprocessing_operation_order': OptionInfo([], "Postprocessing operation order", ui_components.DropdownMulti, lambda: {"choices": [x.name for x in shared_items.postprocessing_scripts()]}), 'upscaling_max_images_in_cache': OptionInfo(5, "Maximum number of images in upscaling cache", gr.Slider, {"minimum": 0, "maximum": 10, "step": 1}), diff --git a/style.css b/style.css index f8b42636d..6e3ca8411 100644 --- a/style.css +++ b/style.css @@ -462,6 +462,15 @@ div.toprow-compact-tools{ padding: 4px; } +#settings > div.tab-nav .settings-category{ + display: block; + margin: 1em 0 0.25em 0; + font-weight: bold; + text-decoration: underline; + cursor: default; + user-select: none; +} + #settings_result{ height: 1.4em; margin: 0 1.2em; From 1f6844eb7e3a91639b2977d1e0cfbb9bf98baea7 Mon Sep 17 00:00:00 2001 From: Jabasukuriputo Wang Date: Sun, 26 Nov 2023 10:04:39 -0600 Subject: [PATCH 085/139] also consider extension url --- modules/ui_extensions.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/modules/ui_extensions.py b/modules/ui_extensions.py index b67088811..252e6ff2c 100644 --- a/modules/ui_extensions.py +++ b/modules/ui_extensions.py @@ -452,6 +452,7 @@ def get_date(info: dict, key): def refresh_available_extensions_from_data(hide_tags, sort_column, filter_text=""): extlist = available_extensions["extensions"] installed_extensions = {extension.name for extension in extensions.extensions} + installed_extension_urls = {normalize_git_url(extension.remote) for extension in extensions.extensions if extension.remote is not None} tags = available_extensions.get("tags", {}) tags_to_hide = set(hide_tags) @@ -484,7 +485,7 @@ def refresh_available_extensions_from_data(hide_tags, sort_column, filter_text=" if url is None: continue - existing = get_extension_dirname_from_url(url) in installed_extensions + existing = get_extension_dirname_from_url(url) in installed_extensions or normalize_git_url(url) in installed_extension_urls extension_tags = extension_tags + ["installed"] if existing else extension_tags if any(x for x in extension_tags if x in tags_to_hide): From b30cc87b786d32f2385cfecf40a2469ee3a96ab5 Mon Sep 17 00:00:00 2001 From: w-e-w <40751091+w-e-w@users.noreply.github.com> Date: Mon, 27 Nov 2023 13:15:17 +0900 Subject: [PATCH 086/139] add Block component creation callback --- modules/gradio_extensons.py | 10 ++++++++++ 1 file changed, 10 insertions(+) diff --git a/modules/gradio_extensons.py b/modules/gradio_extensons.py index e6b6835ad..7d88dc984 100644 --- a/modules/gradio_extensons.py +++ b/modules/gradio_extensons.py @@ -47,10 +47,20 @@ def Block_get_config(self): def BlockContext_init(self, *args, **kwargs): + if scripts.scripts_current is not None: + scripts.scripts_current.before_component(self, **kwargs) + + scripts.script_callbacks.before_component_callback(self, **kwargs) + res = original_BlockContext_init(self, *args, **kwargs) add_classes_to_gradio_component(self) + scripts.script_callbacks.after_component_callback(self, **kwargs) + + if scripts.scripts_current is not None: + scripts.scripts_current.after_component(self, **kwargs) + return res From 8a6e4bda21dddef3ab2e70a05d71b587b6c8b04b Mon Sep 17 00:00:00 2001 From: w-e-w <40751091+w-e-w@users.noreply.github.com> Date: Mon, 27 Nov 2023 14:00:17 +0900 Subject: [PATCH 087/139] catch uncaught exception with ui creation scripts prevent total webui crash --- modules/scripts.py | 54 +++++++++++++++++++++++++--------------------- 1 file changed, 29 insertions(+), 25 deletions(-) diff --git a/modules/scripts.py b/modules/scripts.py index b0689a23d..961d032ce 100644 --- a/modules/scripts.py +++ b/modules/scripts.py @@ -570,40 +570,44 @@ class ScriptRunner: if controls is None: return - script.name = wrap_call(script.title, script.filename, "title", default=script.filename).lower() - api_args = [] + try: + script.name = wrap_call(script.title, script.filename, "title", default=script.filename).lower() + api_args = [] - for control in controls: - control.custom_script_source = os.path.basename(script.filename) + for control in controls: + control.custom_script_source = os.path.basename(script.filename) - arg_info = api_models.ScriptArg(label=control.label or "") + arg_info = api_models.ScriptArg(label=control.label or "") - for field in ("value", "minimum", "maximum", "step"): - v = getattr(control, field, None) - if v is not None: - setattr(arg_info, field, v) + for field in ("value", "minimum", "maximum", "step"): + v = getattr(control, field, None) + if v is not None: + setattr(arg_info, field, v) - choices = getattr(control, 'choices', None) # as of gradio 3.41, some items in choices are strings, and some are tuples where the first elem is the string - if choices is not None: - arg_info.choices = [x[0] if isinstance(x, tuple) else x for x in choices] + choices = getattr(control, 'choices', None) # as of gradio 3.41, some items in choices are strings, and some are tuples where the first elem is the string + if choices is not None: + arg_info.choices = [x[0] if isinstance(x, tuple) else x for x in choices] - api_args.append(arg_info) + api_args.append(arg_info) - script.api_info = api_models.ScriptInfo( - name=script.name, - is_img2img=script.is_img2img, - is_alwayson=script.alwayson, - args=api_args, - ) + script.api_info = api_models.ScriptInfo( + name=script.name, + is_img2img=script.is_img2img, + is_alwayson=script.alwayson, + args=api_args, + ) - if script.infotext_fields is not None: - self.infotext_fields += script.infotext_fields + if script.infotext_fields is not None: + self.infotext_fields += script.infotext_fields - if script.paste_field_names is not None: - self.paste_field_names += script.paste_field_names + if script.paste_field_names is not None: + self.paste_field_names += script.paste_field_names - self.inputs += controls - script.args_to = len(self.inputs) + self.inputs += controls + script.args_to = len(self.inputs) + + except Exception: + errors.report(f"Error creating UI for {script.name}: ", exc_info=True) def setup_ui_for_section(self, section, scriptlist=None): if scriptlist is None: From 9621ca4d64bbe59880d869b923e1572f1475a52b Mon Sep 17 00:00:00 2001 From: Charlie Joynt Date: Mon, 27 Nov 2023 11:39:50 +0000 Subject: [PATCH 088/139] Allow use of mutiple styles csv files --- modules/styles.py | 201 +++++++++++++++++++++++++++++++++++++++------- 1 file changed, 170 insertions(+), 31 deletions(-) diff --git a/modules/styles.py b/modules/styles.py index 0740fe1b1..974d3289b 100644 --- a/modules/styles.py +++ b/modules/styles.py @@ -1,4 +1,5 @@ import csv +import fnmatch import os import os.path import re @@ -10,6 +11,23 @@ class PromptStyle(typing.NamedTuple): name: str prompt: str negative_prompt: str + path: str = None + + +def clean_text(text: str) -> str: + """ + Iterating through a list of regular expressions and replacement strings, we + clean up the prompt and style text to make it easier to match against each + other. + """ + re_list = [ + ("multiple commas", re.compile("(,+\s+)+,?"), ", "), + ("multiple spaces", re.compile("\s{2,}"), " "), + ] + for _, regex, replace in re_list: + text = regex.sub(replace, text) + + return text.strip(", ") def merge_prompts(style_prompt: str, prompt: str) -> str: @@ -26,41 +44,64 @@ def apply_styles_to_prompt(prompt, styles): for style in styles: prompt = merge_prompts(style, prompt) - return prompt + return clean_text(prompt) -re_spaces = re.compile(" +") +def unwrap_style_text_from_prompt(style_text, prompt): + """ + Checks the prompt to see if the style text is wrapped around it. If so, + returns True plus the prompt text without the style text. Otherwise, returns + False with the original prompt. - -def extract_style_text_from_prompt(style_text, prompt): - stripped_prompt = re.sub(re_spaces, " ", prompt.strip()) - stripped_style_text = re.sub(re_spaces, " ", style_text.strip()) + Note that the "cleaned" version of the style text is only used for matching + purposes here. It isn't returned; the original style text is not modified. + """ + stripped_prompt = clean_text(prompt) + stripped_style_text = clean_text(style_text) if "{prompt}" in stripped_style_text: - left, right = stripped_style_text.split("{prompt}", 2) + # Work out whether the prompt is wrapped in the style text. If so, we + # return True and the "inner" prompt text that isn't part of the style. + try: + left, right = stripped_style_text.split("{prompt}", 2) + except ValueError as e: + # If the style text has multple "{prompt}"s, we can't split it into + # two parts. This is an error, but we can't do anything about it. + print("Unable to compare style text to prompt:`n{style_text}") + print(f"Error: {e}") + return False, prompt if stripped_prompt.startswith(left) and stripped_prompt.endswith(right): - prompt = stripped_prompt[len(left):len(stripped_prompt)-len(right)] + prompt = stripped_prompt[len(left) : len(stripped_prompt) - len(right)] return True, prompt else: + # Work out whether the given prompt ends with the style text. If so, we + # return True and the prompt text up to where the style text starts. if stripped_prompt.endswith(stripped_style_text): - prompt = stripped_prompt[:len(stripped_prompt)-len(stripped_style_text)] - - if prompt.endswith(', '): + prompt = stripped_prompt[: len(stripped_prompt) - len(stripped_style_text)] + if prompt.endswith(", "): prompt = prompt[:-2] - return True, prompt return False, prompt -def extract_style_from_prompts(style: PromptStyle, prompt, negative_prompt): +def extract_original_prompts(style: PromptStyle, prompt, negative_prompt): + """ + Takes a style and compares it to the prompt and negative prompt. If the style + matches, returns True plus the prompt and negative prompt with the style text + removed. Otherwise, returns False with the original prompt and negative prompt. + """ if not style.prompt and not style.negative_prompt: return False, prompt, negative_prompt - match_positive, extracted_positive = extract_style_text_from_prompt(style.prompt, prompt) + match_positive, extracted_positive = unwrap_style_text_from_prompt( + style.prompt, prompt + ) if not match_positive: return False, prompt, negative_prompt - match_negative, extracted_negative = extract_style_text_from_prompt(style.negative_prompt, negative_prompt) + match_negative, extracted_negative = unwrap_style_text_from_prompt( + style.negative_prompt, negative_prompt + ) if not match_negative: return False, prompt, negative_prompt @@ -69,25 +110,88 @@ def extract_style_from_prompts(style: PromptStyle, prompt, negative_prompt): class StyleDatabase: def __init__(self, path: str): - self.no_style = PromptStyle("None", "", "") + self.no_style = PromptStyle("None", "", "", None) self.styles = {} self.path = path + folder, file = os.path.split(self.path) + self.default_file = file.split("*")[0] + ".csv" + if self.default_file == ".csv": + self.default_file = "styles.csv" + self.default_path = os.path.join(folder, self.default_file) + + self.prompt_fields = [field for field in PromptStyle._fields if field != "path"] + self.reload() def reload(self): + """ + Clears the style database and reloads the styles from the CSV file(s) + matching the path used to initialize the database. + """ self.styles.clear() - if not os.path.exists(self.path): - return + path, filename = os.path.split(self.path) - with open(self.path, "r", encoding="utf-8-sig", newline='') as file: + if "*" in filename: + fileglob = filename.split("*")[0] + "*.csv" + filelist = [] + for file in os.listdir(path): + if fnmatch.fnmatch(file, fileglob): + filelist.append(file) + # Add a visible divider to the style list + half_len = round(len(file) / 2) + divider = f"{'-' * (20 - half_len)} {file.upper()}" + divider = f"{divider} {'-' * (40 - len(divider))}" + self.styles[divider] = PromptStyle( + f"{divider}", None, None, "do_not_save" + ) + # Add styles from this CSV file + self.load_from_csv(os.path.join(path, file)) + if len(filelist) == 0: + print(f"No styles found in {path} matching {fileglob}") + return + elif not os.path.exists(self.path): + print(f"Style database not found: {self.path}") + return + else: + self.load_from_csv(self.path) + + def load_from_csv(self, path: str): + with open(path, "r", encoding="utf-8-sig", newline="") as file: reader = csv.DictReader(file, skipinitialspace=True) for row in reader: + # Ignore empty rows or rows starting with a comment + if not row or row["name"].startswith("#"): + continue # Support loading old CSV format with "name, text"-columns prompt = row["prompt"] if "prompt" in row else row["text"] negative_prompt = row.get("negative_prompt", "") - self.styles[row["name"]] = PromptStyle(row["name"], prompt, negative_prompt) + # Add style to database + self.styles[row["name"]] = PromptStyle( + row["name"], prompt, negative_prompt, path + ) + + def get_style_paths(self) -> list(): + """ + Returns a list of all distinct paths, including the default path, of + files that styles are loaded from.""" + # Update any styles without a path to the default path + for style in list(self.styles.values()): + if not style.path: + self.styles[style.name] = style._replace(path=self.default_path) + + # Create a list of all distinct paths, including the default path + style_paths = set() + style_paths.add(self.default_path) + for _, style in self.styles.items(): + if style.path: + style_paths.add(style.path) + + # Remove any paths for styles that are just list dividers + style_paths.remove("do_not_save") + + return list(style_paths) def get_style_prompts(self, styles): return [self.styles.get(x, self.no_style).prompt for x in styles] @@ -96,20 +200,53 @@ class StyleDatabase: return [self.styles.get(x, self.no_style).negative_prompt for x in styles] def apply_styles_to_prompt(self, prompt, styles): - return apply_styles_to_prompt(prompt, [self.styles.get(x, self.no_style).prompt for x in styles]) + return apply_styles_to_prompt( + prompt, [self.styles.get(x, self.no_style).prompt for x in styles] + ) def apply_negative_styles_to_prompt(self, prompt, styles): - return apply_styles_to_prompt(prompt, [self.styles.get(x, self.no_style).negative_prompt for x in styles]) + return apply_styles_to_prompt( + prompt, [self.styles.get(x, self.no_style).negative_prompt for x in styles] + ) - def save_styles(self, path: str) -> None: - # Always keep a backup file around - if os.path.exists(path): - shutil.copy(path, f"{path}.bak") + def save_styles(self, path: str = None) -> None: + # The path argument is deprecated, but kept for backwards compatibility + _ = path - with open(path, "w", encoding="utf-8-sig", newline='') as file: - writer = csv.DictWriter(file, fieldnames=PromptStyle._fields) - writer.writeheader() - writer.writerows(style._asdict() for k, style in self.styles.items()) + # Update any styles without a path to the default path + for style in list(self.styles.values()): + if not style.path: + self.styles[style.name] = style._replace(path=self.default_path) + + # Create a list of all distinct paths, including the default path + style_paths = set() + style_paths.add(self.default_path) + for _, style in self.styles.items(): + if style.path: + style_paths.add(style.path) + + # Remove any paths for styles that are just list dividers + style_paths.remove("do_not_save") + + csv_names = [os.path.split(path)[1].lower() for path in style_paths] + + for style_path in style_paths: + # Always keep a backup file around + if os.path.exists(style_path): + shutil.copy(style_path, f"{style_path}.bak") + + # Write the styles to the CSV file + with open(style_path, "w", encoding="utf-8-sig", newline="") as file: + writer = csv.DictWriter(file, fieldnames=self.prompt_fields) + writer.writeheader() + for style in (s for s in self.styles.values() if s.path == style_path): + # Skip style list dividers, e.g. "STYLES.CSV" + if style.name.lower().strip("# ") in csv_names: + continue + # Write style fields, ignoring the path field + writer.writerow( + {k: v for k, v in style._asdict().items() if k != "path"} + ) def extract_styles_from_prompt(self, prompt, negative_prompt): extracted = [] @@ -120,7 +257,9 @@ class StyleDatabase: found_style = None for style in applicable_styles: - is_match, new_prompt, new_neg_prompt = extract_style_from_prompts(style, prompt, negative_prompt) + is_match, new_prompt, new_neg_prompt = extract_original_prompts( + style, prompt, negative_prompt + ) if is_match: found_style = style prompt = new_prompt From 1c64bb71402c2cd62ac98f936203437f0c4fcd02 Mon Sep 17 00:00:00 2001 From: MisterSeajay Date: Mon, 27 Nov 2023 11:57:27 +0000 Subject: [PATCH 089/139] bugfix for warning message (#6) --- modules/styles.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/modules/styles.py b/modules/styles.py index 974d3289b..e73920c70 100644 --- a/modules/styles.py +++ b/modules/styles.py @@ -66,7 +66,7 @@ def unwrap_style_text_from_prompt(style_text, prompt): except ValueError as e: # If the style text has multple "{prompt}"s, we can't split it into # two parts. This is an error, but we can't do anything about it. - print("Unable to compare style text to prompt:`n{style_text}") + print(f"Unable to compare style text to prompt:`n{style_text}") print(f"Error: {e}") return False, prompt if stripped_prompt.startswith(left) and stripped_prompt.endswith(right): From a75314b41f938d1e598916ecdd0f14126ae1876b Mon Sep 17 00:00:00 2001 From: MisterSeajay Date: Mon, 27 Nov 2023 12:03:42 +0000 Subject: [PATCH 090/139] bugfix for warning message (#6) * bugfix for warning message * bugfix error message --- modules/styles.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/modules/styles.py b/modules/styles.py index e73920c70..4d218cd7e 100644 --- a/modules/styles.py +++ b/modules/styles.py @@ -66,7 +66,7 @@ def unwrap_style_text_from_prompt(style_text, prompt): except ValueError as e: # If the style text has multple "{prompt}"s, we can't split it into # two parts. This is an error, but we can't do anything about it. - print(f"Unable to compare style text to prompt:`n{style_text}") + print(f"Unable to compare style text to prompt:\n{style_text}") print(f"Error: {e}") return False, prompt if stripped_prompt.startswith(left) and stripped_prompt.endswith(right): From 26a0c29587da428d27fd3e6a95491776ef66bbdd Mon Sep 17 00:00:00 2001 From: Charlie Joynt Date: Mon, 27 Nov 2023 11:39:50 +0000 Subject: [PATCH 091/139] Allow use of mutiple styles csv files * https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/14122 Fix edge case where style text has multiple {prompt} placeholders * https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/14005 --- modules/styles.py | 201 +++++++++++++++++++++++++++++++++++++++------- 1 file changed, 170 insertions(+), 31 deletions(-) diff --git a/modules/styles.py b/modules/styles.py index 0740fe1b1..4d218cd7e 100644 --- a/modules/styles.py +++ b/modules/styles.py @@ -1,4 +1,5 @@ import csv +import fnmatch import os import os.path import re @@ -10,6 +11,23 @@ class PromptStyle(typing.NamedTuple): name: str prompt: str negative_prompt: str + path: str = None + + +def clean_text(text: str) -> str: + """ + Iterating through a list of regular expressions and replacement strings, we + clean up the prompt and style text to make it easier to match against each + other. + """ + re_list = [ + ("multiple commas", re.compile("(,+\s+)+,?"), ", "), + ("multiple spaces", re.compile("\s{2,}"), " "), + ] + for _, regex, replace in re_list: + text = regex.sub(replace, text) + + return text.strip(", ") def merge_prompts(style_prompt: str, prompt: str) -> str: @@ -26,41 +44,64 @@ def apply_styles_to_prompt(prompt, styles): for style in styles: prompt = merge_prompts(style, prompt) - return prompt + return clean_text(prompt) -re_spaces = re.compile(" +") +def unwrap_style_text_from_prompt(style_text, prompt): + """ + Checks the prompt to see if the style text is wrapped around it. If so, + returns True plus the prompt text without the style text. Otherwise, returns + False with the original prompt. - -def extract_style_text_from_prompt(style_text, prompt): - stripped_prompt = re.sub(re_spaces, " ", prompt.strip()) - stripped_style_text = re.sub(re_spaces, " ", style_text.strip()) + Note that the "cleaned" version of the style text is only used for matching + purposes here. It isn't returned; the original style text is not modified. + """ + stripped_prompt = clean_text(prompt) + stripped_style_text = clean_text(style_text) if "{prompt}" in stripped_style_text: - left, right = stripped_style_text.split("{prompt}", 2) + # Work out whether the prompt is wrapped in the style text. If so, we + # return True and the "inner" prompt text that isn't part of the style. + try: + left, right = stripped_style_text.split("{prompt}", 2) + except ValueError as e: + # If the style text has multple "{prompt}"s, we can't split it into + # two parts. This is an error, but we can't do anything about it. + print(f"Unable to compare style text to prompt:\n{style_text}") + print(f"Error: {e}") + return False, prompt if stripped_prompt.startswith(left) and stripped_prompt.endswith(right): - prompt = stripped_prompt[len(left):len(stripped_prompt)-len(right)] + prompt = stripped_prompt[len(left) : len(stripped_prompt) - len(right)] return True, prompt else: + # Work out whether the given prompt ends with the style text. If so, we + # return True and the prompt text up to where the style text starts. if stripped_prompt.endswith(stripped_style_text): - prompt = stripped_prompt[:len(stripped_prompt)-len(stripped_style_text)] - - if prompt.endswith(', '): + prompt = stripped_prompt[: len(stripped_prompt) - len(stripped_style_text)] + if prompt.endswith(", "): prompt = prompt[:-2] - return True, prompt return False, prompt -def extract_style_from_prompts(style: PromptStyle, prompt, negative_prompt): +def extract_original_prompts(style: PromptStyle, prompt, negative_prompt): + """ + Takes a style and compares it to the prompt and negative prompt. If the style + matches, returns True plus the prompt and negative prompt with the style text + removed. Otherwise, returns False with the original prompt and negative prompt. + """ if not style.prompt and not style.negative_prompt: return False, prompt, negative_prompt - match_positive, extracted_positive = extract_style_text_from_prompt(style.prompt, prompt) + match_positive, extracted_positive = unwrap_style_text_from_prompt( + style.prompt, prompt + ) if not match_positive: return False, prompt, negative_prompt - match_negative, extracted_negative = extract_style_text_from_prompt(style.negative_prompt, negative_prompt) + match_negative, extracted_negative = unwrap_style_text_from_prompt( + style.negative_prompt, negative_prompt + ) if not match_negative: return False, prompt, negative_prompt @@ -69,25 +110,88 @@ def extract_style_from_prompts(style: PromptStyle, prompt, negative_prompt): class StyleDatabase: def __init__(self, path: str): - self.no_style = PromptStyle("None", "", "") + self.no_style = PromptStyle("None", "", "", None) self.styles = {} self.path = path + folder, file = os.path.split(self.path) + self.default_file = file.split("*")[0] + ".csv" + if self.default_file == ".csv": + self.default_file = "styles.csv" + self.default_path = os.path.join(folder, self.default_file) + + self.prompt_fields = [field for field in PromptStyle._fields if field != "path"] + self.reload() def reload(self): + """ + Clears the style database and reloads the styles from the CSV file(s) + matching the path used to initialize the database. + """ self.styles.clear() - if not os.path.exists(self.path): - return + path, filename = os.path.split(self.path) - with open(self.path, "r", encoding="utf-8-sig", newline='') as file: + if "*" in filename: + fileglob = filename.split("*")[0] + "*.csv" + filelist = [] + for file in os.listdir(path): + if fnmatch.fnmatch(file, fileglob): + filelist.append(file) + # Add a visible divider to the style list + half_len = round(len(file) / 2) + divider = f"{'-' * (20 - half_len)} {file.upper()}" + divider = f"{divider} {'-' * (40 - len(divider))}" + self.styles[divider] = PromptStyle( + f"{divider}", None, None, "do_not_save" + ) + # Add styles from this CSV file + self.load_from_csv(os.path.join(path, file)) + if len(filelist) == 0: + print(f"No styles found in {path} matching {fileglob}") + return + elif not os.path.exists(self.path): + print(f"Style database not found: {self.path}") + return + else: + self.load_from_csv(self.path) + + def load_from_csv(self, path: str): + with open(path, "r", encoding="utf-8-sig", newline="") as file: reader = csv.DictReader(file, skipinitialspace=True) for row in reader: + # Ignore empty rows or rows starting with a comment + if not row or row["name"].startswith("#"): + continue # Support loading old CSV format with "name, text"-columns prompt = row["prompt"] if "prompt" in row else row["text"] negative_prompt = row.get("negative_prompt", "") - self.styles[row["name"]] = PromptStyle(row["name"], prompt, negative_prompt) + # Add style to database + self.styles[row["name"]] = PromptStyle( + row["name"], prompt, negative_prompt, path + ) + + def get_style_paths(self) -> list(): + """ + Returns a list of all distinct paths, including the default path, of + files that styles are loaded from.""" + # Update any styles without a path to the default path + for style in list(self.styles.values()): + if not style.path: + self.styles[style.name] = style._replace(path=self.default_path) + + # Create a list of all distinct paths, including the default path + style_paths = set() + style_paths.add(self.default_path) + for _, style in self.styles.items(): + if style.path: + style_paths.add(style.path) + + # Remove any paths for styles that are just list dividers + style_paths.remove("do_not_save") + + return list(style_paths) def get_style_prompts(self, styles): return [self.styles.get(x, self.no_style).prompt for x in styles] @@ -96,20 +200,53 @@ class StyleDatabase: return [self.styles.get(x, self.no_style).negative_prompt for x in styles] def apply_styles_to_prompt(self, prompt, styles): - return apply_styles_to_prompt(prompt, [self.styles.get(x, self.no_style).prompt for x in styles]) + return apply_styles_to_prompt( + prompt, [self.styles.get(x, self.no_style).prompt for x in styles] + ) def apply_negative_styles_to_prompt(self, prompt, styles): - return apply_styles_to_prompt(prompt, [self.styles.get(x, self.no_style).negative_prompt for x in styles]) + return apply_styles_to_prompt( + prompt, [self.styles.get(x, self.no_style).negative_prompt for x in styles] + ) - def save_styles(self, path: str) -> None: - # Always keep a backup file around - if os.path.exists(path): - shutil.copy(path, f"{path}.bak") + def save_styles(self, path: str = None) -> None: + # The path argument is deprecated, but kept for backwards compatibility + _ = path - with open(path, "w", encoding="utf-8-sig", newline='') as file: - writer = csv.DictWriter(file, fieldnames=PromptStyle._fields) - writer.writeheader() - writer.writerows(style._asdict() for k, style in self.styles.items()) + # Update any styles without a path to the default path + for style in list(self.styles.values()): + if not style.path: + self.styles[style.name] = style._replace(path=self.default_path) + + # Create a list of all distinct paths, including the default path + style_paths = set() + style_paths.add(self.default_path) + for _, style in self.styles.items(): + if style.path: + style_paths.add(style.path) + + # Remove any paths for styles that are just list dividers + style_paths.remove("do_not_save") + + csv_names = [os.path.split(path)[1].lower() for path in style_paths] + + for style_path in style_paths: + # Always keep a backup file around + if os.path.exists(style_path): + shutil.copy(style_path, f"{style_path}.bak") + + # Write the styles to the CSV file + with open(style_path, "w", encoding="utf-8-sig", newline="") as file: + writer = csv.DictWriter(file, fieldnames=self.prompt_fields) + writer.writeheader() + for style in (s for s in self.styles.values() if s.path == style_path): + # Skip style list dividers, e.g. "STYLES.CSV" + if style.name.lower().strip("# ") in csv_names: + continue + # Write style fields, ignoring the path field + writer.writerow( + {k: v for k, v in style._asdict().items() if k != "path"} + ) def extract_styles_from_prompt(self, prompt, negative_prompt): extracted = [] @@ -120,7 +257,9 @@ class StyleDatabase: found_style = None for style in applicable_styles: - is_match, new_prompt, new_neg_prompt = extract_style_from_prompts(style, prompt, negative_prompt) + is_match, new_prompt, new_neg_prompt = extract_original_prompts( + style, prompt, negative_prompt + ) if is_match: found_style = style prompt = new_prompt From 23c36f59b4a423362d74f1ca2cc69871ae101e0e Mon Sep 17 00:00:00 2001 From: aria1th <35677394+aria1th@users.noreply.github.com> Date: Mon, 27 Nov 2023 21:10:26 +0900 Subject: [PATCH 092/139] Support XYZ scripts / split hires path from unet --- .../hypertile/scripts/hypertile_script.py | 11 ++-- .../hypertile/scripts/hypertile_xyz.py | 52 +++++++++++++++++++ 2 files changed, 60 insertions(+), 3 deletions(-) create mode 100644 extensions-builtin/hypertile/scripts/hypertile_xyz.py diff --git a/extensions-builtin/hypertile/scripts/hypertile_script.py b/extensions-builtin/hypertile/scripts/hypertile_script.py index 3cc29cd1f..b2413cc5f 100644 --- a/extensions-builtin/hypertile/scripts/hypertile_script.py +++ b/extensions-builtin/hypertile/scripts/hypertile_script.py @@ -1,5 +1,6 @@ import hypertile from modules import scripts, script_callbacks, shared +from scripts.hypertile_xyz import add_axis_options class ScriptHypertile(scripts.Script): @@ -17,7 +18,10 @@ class ScriptHypertile(scripts.Script): configure_hypertile(p.width, p.height, enable_unet=shared.opts.hypertile_enable_unet) def before_hr(self, p, *args): - configure_hypertile(p.hr_upscale_to_x, p.hr_upscale_to_y, enable_unet=shared.opts.hypertile_enable_unet_secondpass or shared.opts.hypertile_enable_unet) + # exclusive hypertile seed for the second pass + if not shared.opts.hypertile_enable_unet: + hypertile.set_hypertile_seed(p.all_seeds[0]) + configure_hypertile(p.hr_upscale_to_x, p.hr_upscale_to_y, enable_unet=shared.opts.hypertile_enable_unet_secondpass) def configure_hypertile(width, height, enable_unet=True): @@ -57,12 +61,12 @@ def on_ui_settings(): "hypertile_enable_unet_secondpass": shared.OptionInfo(False, "Enable Hypertile U-Net for hires fix second pass"), "hypertile_max_depth_unet": shared.OptionInfo(3, "Hypertile U-Net max depth", gr.Slider, {"minimum": 0, "maximum": 3, "step": 1}), "hypertile_max_tile_unet": shared.OptionInfo(256, "Hypertile U-net max tile size", gr.Slider, {"minimum": 0, "maximum": 512, "step": 16}), - "hypertile_swap_size_unet": shared.OptionInfo(3, "Hypertile U-net swap size", gr.Slider, {"minimum": 0, "maximum": 6, "step": 1}), + "hypertile_swap_size_unet": shared.OptionInfo(3, "Hypertile U-net swap size", gr.Slider, {"minimum": 0, "maximum": 64, "step": 1}), "hypertile_enable_vae": shared.OptionInfo(False, "Enable Hypertile VAE").info("minimal change in the generated picture"), "hypertile_max_depth_vae": shared.OptionInfo(3, "Hypertile VAE max depth", gr.Slider, {"minimum": 0, "maximum": 3, "step": 1}), "hypertile_max_tile_vae": shared.OptionInfo(128, "Hypertile VAE max tile size", gr.Slider, {"minimum": 0, "maximum": 512, "step": 16}), - "hypertile_swap_size_vae": shared.OptionInfo(3, "Hypertile VAE swap size ", gr.Slider, {"minimum": 0, "maximum": 6, "step": 1}), + "hypertile_swap_size_vae": shared.OptionInfo(3, "Hypertile VAE swap size ", gr.Slider, {"minimum": 0, "maximum": 64, "step": 1}), } for name, opt in options.items(): @@ -71,3 +75,4 @@ def on_ui_settings(): script_callbacks.on_ui_settings(on_ui_settings) +script_callbacks.on_before_ui(add_axis_options) \ No newline at end of file diff --git a/extensions-builtin/hypertile/scripts/hypertile_xyz.py b/extensions-builtin/hypertile/scripts/hypertile_xyz.py new file mode 100644 index 000000000..eaf7c8d7a --- /dev/null +++ b/extensions-builtin/hypertile/scripts/hypertile_xyz.py @@ -0,0 +1,52 @@ +from modules import scripts +xyz_grid = [x for x in scripts.scripts_data if x.script_class.__module__ == "xyz_grid.py"][0].module +from modules.shared import opts + +def int_applier(value_name:str, min_range:int = -1, max_range:int = -1): + """ + Returns a function that applies the given value to the given value_name in opts.data. + """ + # convert to int + def validate(value_name:str, value:str): + try: + value = int(value) + except: + raise ValueError(f"Value {value} for {value_name} is not an integer") + # validate value + if not min_range == -1: + assert value >= min_range, f"Value {value} for {value_name} must be greater than or equal to {min_range}" + if not max_range == -1: + assert value <= max_range, f"Value {value} for {value_name} must be less than or equal to {max_range}" + def apply_int(p, x, xs): + validate(value_name, x) + opts.data[value_name] = int(x) + return apply_int + +def bool_applier(value_name:str): + """ + Returns a function that applies the given value to the given value_name in opts.data. + """ + def validate(value_name:str, value:str): + assert value.lower() in ["true", "false"], f"Value {value} for {value_name} must be either true or false" + def apply_bool(p, x, xs): + validate(value_name, x) + value_boolean = x.lower() == "true" + opts.data[value_name] = value_boolean + return apply_bool + +def add_axis_options(): + extra_axis_options = [ + xyz_grid.AxisOption("[Hypertile] Unet First pass Enabled", str, bool_applier("hypertile_enable_unet"), choices=xyz_grid.boolean_choice(reverse=True)), + xyz_grid.AxisOption("[Hypertile] Unet Second pass Enabled", str, bool_applier("hypertile_enable_unet_secondpass"), choices=xyz_grid.boolean_choice(reverse=True)), + xyz_grid.AxisOption("[Hypertile] Unet Max Depth", int, int_applier("hypertile_max_depth_unet", 0, 3), choices=lambda: [str(x) for x in range(4)]), + xyz_grid.AxisOption("[Hypertile] Unet Max Tile Size", int, int_applier("hypertile_max_tile_unet", 0, 512)), + xyz_grid.AxisOption("[Hypertile] Unet Swap Size", int, int_applier("hypertile_swap_size_unet", 0, 64)), + xyz_grid.AxisOption("[Hypertile] VAE Enabled", str, bool_applier("hypertile_enable_vae"), choices=xyz_grid.boolean_choice(reverse=True)), + xyz_grid.AxisOption("[Hypertile] VAE Max Depth", int, int_applier("hypertile_max_depth_vae", 0, 3), choices=lambda: [str(x) for x in range(4)]), + xyz_grid.AxisOption("[Hypertile] VAE Max Tile Size", int, int_applier("hypertile_max_tile_vae", 0, 512)), + xyz_grid.AxisOption("[Hypertile] VAE Swap Size", int, int_applier("hypertile_swap_size_vae", 0, 64)), + ] + # check if the axis options have already been added + if any(set(opt.label for opt in extra_axis_options).intersection(set(opt.label for opt in xyz_grid.axis_options))): + return + xyz_grid.axis_options.extend(extra_axis_options) \ No newline at end of file From 601a7b4ce5b28efd29b1668c7b8b74fb6b62f6f3 Mon Sep 17 00:00:00 2001 From: aria1th <35677394+aria1th@users.noreply.github.com> Date: Mon, 27 Nov 2023 22:10:31 +0900 Subject: [PATCH 093/139] cache divisors / fix ruff --- extensions-builtin/hypertile/hypertile.py | 24 ++++++++++++------- .../hypertile/scripts/hypertile_script.py | 2 +- .../hypertile/scripts/hypertile_xyz.py | 18 +++++++------- 3 files changed, 26 insertions(+), 18 deletions(-) diff --git a/extensions-builtin/hypertile/hypertile.py b/extensions-builtin/hypertile/hypertile.py index feb02fd27..0f40e2d39 100644 --- a/extensions-builtin/hypertile/hypertile.py +++ b/extensions-builtin/hypertile/hypertile.py @@ -6,7 +6,6 @@ Original author: @tfernd Github: https://github.com/tfernd/HyperTile from __future__ import annotations -import functools from dataclasses import dataclass from typing import Callable @@ -189,6 +188,19 @@ DEPTH_LAYERS_XL = { RNG_INSTANCE = random.Random() +@cache +def get_divisors(value: int, min_value: int, /, max_options: int = 1) -> list[int]: + """ + Returns divisors of value that + x * min_value <= value + in big -> small order, amount of divisors is limited by max_options + """ + max_options = max(1, max_options) # at least 1 option should be returned + min_value = min(min_value, value) + divisors = [i for i in range(min_value, value + 1) if value % i == 0] # divisors in small -> big order + ns = [value // i for i in divisors[:max_options]] # has at least 1 element # big -> small order + return ns + def random_divisor(value: int, min_value: int, /, max_options: int = 1) -> int: """ @@ -196,13 +208,7 @@ def random_divisor(value: int, min_value: int, /, max_options: int = 1) -> int: x * min_value <= value if max_options is 1, the behavior is deterministic """ - min_value = min(min_value, value) - - # All big divisors of value (inclusive) - divisors = [i for i in range(min_value, value + 1) if value % i == 0] # divisors in small -> big order - - ns = [value // i for i in divisors[:max_options]] # has at least 1 element # big -> small order - + ns = get_divisors(value, min_value, max_options=max_options) # get cached divisors idx = RNG_INSTANCE.randint(0, len(ns) - 1) return ns[idx] @@ -212,7 +218,7 @@ def set_hypertile_seed(seed: int) -> None: RNG_INSTANCE.seed(seed) -@functools.cache +@cache def largest_tile_size_available(width: int, height: int) -> int: """ Calculates the largest tile size available for a given width and height diff --git a/extensions-builtin/hypertile/scripts/hypertile_script.py b/extensions-builtin/hypertile/scripts/hypertile_script.py index b2413cc5f..d3ab60915 100644 --- a/extensions-builtin/hypertile/scripts/hypertile_script.py +++ b/extensions-builtin/hypertile/scripts/hypertile_script.py @@ -75,4 +75,4 @@ def on_ui_settings(): script_callbacks.on_ui_settings(on_ui_settings) -script_callbacks.on_before_ui(add_axis_options) \ No newline at end of file +script_callbacks.on_before_ui(add_axis_options) diff --git a/extensions-builtin/hypertile/scripts/hypertile_xyz.py b/extensions-builtin/hypertile/scripts/hypertile_xyz.py index eaf7c8d7a..3007a0832 100644 --- a/extensions-builtin/hypertile/scripts/hypertile_xyz.py +++ b/extensions-builtin/hypertile/scripts/hypertile_xyz.py @@ -1,17 +1,17 @@ from modules import scripts -xyz_grid = [x for x in scripts.scripts_data if x.script_class.__module__ == "xyz_grid.py"][0].module from modules.shared import opts +xyz_grid = [x for x in scripts.scripts_data if x.script_class.__module__ == "xyz_grid.py"][0].module + def int_applier(value_name:str, min_range:int = -1, max_range:int = -1): """ Returns a function that applies the given value to the given value_name in opts.data. """ # convert to int def validate(value_name:str, value:str): - try: - value = int(value) - except: - raise ValueError(f"Value {value} for {value_name} is not an integer") + if not value.isnumeric(): + raise ValueError(f"Value {value} for {value_name} must be an integer") + value = int(value) # validate value if not min_range == -1: assert value >= min_range, f"Value {value} for {value_name} must be greater than or equal to {min_range}" @@ -46,7 +46,9 @@ def add_axis_options(): xyz_grid.AxisOption("[Hypertile] VAE Max Tile Size", int, int_applier("hypertile_max_tile_vae", 0, 512)), xyz_grid.AxisOption("[Hypertile] VAE Swap Size", int, int_applier("hypertile_swap_size_vae", 0, 64)), ] - # check if the axis options have already been added - if any(set(opt.label for opt in extra_axis_options).intersection(set(opt.label for opt in xyz_grid.axis_options))): + set_a = set([opt.label for opt in xyz_grid.axis_options]) + set_b = set([opt.label for opt in extra_axis_options]) + if set_a.intersection(set_b): return - xyz_grid.axis_options.extend(extra_axis_options) \ No newline at end of file + + xyz_grid.axis_options.extend(extra_axis_options) From f207eb7a0d8b4443dbe665df99c31f8ff91660fd Mon Sep 17 00:00:00 2001 From: aria1th <35677394+aria1th@users.noreply.github.com> Date: Mon, 27 Nov 2023 22:11:28 +0900 Subject: [PATCH 094/139] fix ruff in hypertile_xyz.py --- extensions-builtin/hypertile/scripts/hypertile_xyz.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/extensions-builtin/hypertile/scripts/hypertile_xyz.py b/extensions-builtin/hypertile/scripts/hypertile_xyz.py index 3007a0832..4055a9ead 100644 --- a/extensions-builtin/hypertile/scripts/hypertile_xyz.py +++ b/extensions-builtin/hypertile/scripts/hypertile_xyz.py @@ -46,8 +46,8 @@ def add_axis_options(): xyz_grid.AxisOption("[Hypertile] VAE Max Tile Size", int, int_applier("hypertile_max_tile_vae", 0, 512)), xyz_grid.AxisOption("[Hypertile] VAE Swap Size", int, int_applier("hypertile_swap_size_vae", 0, 64)), ] - set_a = set([opt.label for opt in xyz_grid.axis_options]) - set_b = set([opt.label for opt in extra_axis_options]) + set_a = set(opt.label for opt in xyz_grid.axis_options) + set_b = set(opt.label for opt in extra_axis_options) if set_a.intersection(set_b): return From 524d6a4dbae68bf557d9c5fe686707d96841e0b5 Mon Sep 17 00:00:00 2001 From: aria1th <35677394+aria1th@users.noreply.github.com> Date: Mon, 27 Nov 2023 22:13:18 +0900 Subject: [PATCH 095/139] fix ruff - set comprehension --- extensions-builtin/hypertile/scripts/hypertile_xyz.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/extensions-builtin/hypertile/scripts/hypertile_xyz.py b/extensions-builtin/hypertile/scripts/hypertile_xyz.py index 4055a9ead..928e99652 100644 --- a/extensions-builtin/hypertile/scripts/hypertile_xyz.py +++ b/extensions-builtin/hypertile/scripts/hypertile_xyz.py @@ -46,8 +46,8 @@ def add_axis_options(): xyz_grid.AxisOption("[Hypertile] VAE Max Tile Size", int, int_applier("hypertile_max_tile_vae", 0, 512)), xyz_grid.AxisOption("[Hypertile] VAE Swap Size", int, int_applier("hypertile_swap_size_vae", 0, 64)), ] - set_a = set(opt.label for opt in xyz_grid.axis_options) - set_b = set(opt.label for opt in extra_axis_options) + set_a = {opt.label for opt in xyz_grid.axis_options} + set_b = {opt.label for opt in extra_axis_options} if set_a.intersection(set_b): return From ec78354efa179b64e92d6b98d781f6572b4eb084 Mon Sep 17 00:00:00 2001 From: aria1th <35677394+aria1th@users.noreply.github.com> Date: Mon, 27 Nov 2023 22:25:28 +0900 Subject: [PATCH 096/139] hypertile_xyz: we don't need isnumeric check for AxisOption --- extensions-builtin/hypertile/scripts/hypertile_xyz.py | 3 --- 1 file changed, 3 deletions(-) diff --git a/extensions-builtin/hypertile/scripts/hypertile_xyz.py b/extensions-builtin/hypertile/scripts/hypertile_xyz.py index 928e99652..9e96ae3c5 100644 --- a/extensions-builtin/hypertile/scripts/hypertile_xyz.py +++ b/extensions-builtin/hypertile/scripts/hypertile_xyz.py @@ -7,10 +7,7 @@ def int_applier(value_name:str, min_range:int = -1, max_range:int = -1): """ Returns a function that applies the given value to the given value_name in opts.data. """ - # convert to int def validate(value_name:str, value:str): - if not value.isnumeric(): - raise ValueError(f"Value {value} for {value_name} must be an integer") value = int(value) # validate value if not min_range == -1: From 3cd6e1d0a0877e6f1ac931c8253e6eee09da3805 Mon Sep 17 00:00:00 2001 From: obsol <33932119+read-0nly@users.noreply.github.com> Date: Mon, 27 Nov 2023 19:21:43 -0500 Subject: [PATCH 097/139] Update devices.py fixes issue where "--use-cpu" all properly makes SD run on CPU but leaves ControlNet (and other extensions, I presume) pointed at GPU, causing a crash in ControlNet caused by a mismatch between devices between SD and CN https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/14097 --- modules/devices.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/modules/devices.py b/modules/devices.py index c01f06024..65efcf1eb 100644 --- a/modules/devices.py +++ b/modules/devices.py @@ -38,7 +38,7 @@ def get_optimal_device(): def get_device_for(task): - if task in shared.cmd_opts.use_cpu: + if task in shared.cmd_opts.use_cpu or "all" in shared.cmd_opts.use_cpu: return cpu return get_optimal_device() From 03ee297aa22296ea12b965fc1cb11aa46375d372 Mon Sep 17 00:00:00 2001 From: w-e-w <40751091+w-e-w@users.noreply.github.com> Date: Mon, 27 Nov 2023 17:26:16 +0900 Subject: [PATCH 098/139] fix Auto focal point crop for opencv >= 4.8.x autocrop.download_and_cache_models in opencv >= 4.8 the face detection model was updated download the base on opencv version returns the model path or raise exception --- modules/textual_inversion/autocrop.py | 29 ++++++++++++++----------- modules/textual_inversion/preprocess.py | 4 ++-- 2 files changed, 18 insertions(+), 15 deletions(-) diff --git a/modules/textual_inversion/autocrop.py b/modules/textual_inversion/autocrop.py index 1675e39a5..051be1188 100644 --- a/modules/textual_inversion/autocrop.py +++ b/modules/textual_inversion/autocrop.py @@ -3,6 +3,8 @@ import requests import os import numpy as np from PIL import ImageDraw +from modules import paths_internal +from pkg_resources import parse_version GREEN = "#0F0" BLUE = "#00F" @@ -294,22 +296,23 @@ def is_square(w, h): return w == h -def download_and_cache_models(dirname): - download_url = 'https://github.com/opencv/opencv_zoo/blob/91fb0290f50896f38a0ab1e558b74b16bc009428/models/face_detection_yunet/face_detection_yunet_2022mar.onnx?raw=true' - model_file_name = 'face_detection_yunet.onnx' +model_dir_opencv = os.path.join(paths_internal.models_path, 'opencv') +if parse_version(cv2.__version__) >= parse_version('4.8'): + model_file_path = os.path.join(model_dir_opencv, 'face_detection_yunet_2023mar.onnx') + model_url = 'https://github.com/opencv/opencv_zoo/blob/b6e370b10f641879a87890d44e42173077154a05/models/face_detection_yunet/face_detection_yunet_2023mar.onnx?raw=true' +else: + model_file_path = os.path.join(model_dir_opencv, 'face_detection_yunet.onnx') + model_url = 'https://github.com/opencv/opencv_zoo/blob/91fb0290f50896f38a0ab1e558b74b16bc009428/models/face_detection_yunet/face_detection_yunet_2022mar.onnx?raw=true' - os.makedirs(dirname, exist_ok=True) - cache_file = os.path.join(dirname, model_file_name) - if not os.path.exists(cache_file): - print(f"downloading face detection model from '{download_url}' to '{cache_file}'") - response = requests.get(download_url) - with open(cache_file, "wb") as f: +def download_and_cache_models(): + if not os.path.exists(model_file_path): + os.makedirs(model_dir_opencv, exist_ok=True) + print(f"downloading face detection model from '{model_url}' to '{model_file_path}'") + response = requests.get(model_url) + with open(model_file_path, "wb") as f: f.write(response.content) - - if os.path.exists(cache_file): - return cache_file - return None + return model_file_path class PointOfInterest: diff --git a/modules/textual_inversion/preprocess.py b/modules/textual_inversion/preprocess.py index dbd856bd8..789fa0838 100644 --- a/modules/textual_inversion/preprocess.py +++ b/modules/textual_inversion/preprocess.py @@ -3,7 +3,7 @@ from PIL import Image, ImageOps import math import tqdm -from modules import paths, shared, images, deepbooru +from modules import shared, images, deepbooru from modules.textual_inversion import autocrop @@ -196,7 +196,7 @@ def preprocess_work(process_src, process_dst, process_width, process_height, pre dnn_model_path = None try: - dnn_model_path = autocrop.download_and_cache_models(os.path.join(paths.models_path, "opencv")) + dnn_model_path = autocrop.download_and_cache_models() except Exception as e: print("Unable to load face detection model for auto crop selection. Falling back to lower quality haar method.", e) From d608926f817b279d16b39a7875beec80d010a988 Mon Sep 17 00:00:00 2001 From: w-e-w <40751091+w-e-w@users.noreply.github.com> Date: Tue, 28 Nov 2023 12:12:27 +0900 Subject: [PATCH 099/139] reformat file with uniform indentation --- modules/textual_inversion/autocrop.py | 204 +++++++++++++------------- 1 file changed, 103 insertions(+), 101 deletions(-) diff --git a/modules/textual_inversion/autocrop.py b/modules/textual_inversion/autocrop.py index 051be1188..e223a2e0c 100644 --- a/modules/textual_inversion/autocrop.py +++ b/modules/textual_inversion/autocrop.py @@ -27,7 +27,6 @@ def crop_image(im, settings): elif is_portrait(settings.crop_width, settings.crop_height): scale_by = settings.crop_height / im.height - im = im.resize((int(im.width * scale_by), int(im.height * scale_by))) im_debug = im.copy() @@ -71,6 +70,7 @@ def crop_image(im, settings): return results + def focal_point(im, settings): corner_points = image_corner_points(im, settings) if settings.corner_points_weight > 0 else [] entropy_points = image_entropy_points(im, settings) if settings.entropy_points_weight > 0 else [] @@ -80,118 +80,120 @@ def focal_point(im, settings): weight_pref_total = 0 if corner_points: - weight_pref_total += settings.corner_points_weight + weight_pref_total += settings.corner_points_weight if entropy_points: - weight_pref_total += settings.entropy_points_weight + weight_pref_total += settings.entropy_points_weight if face_points: - weight_pref_total += settings.face_points_weight + weight_pref_total += settings.face_points_weight corner_centroid = None if corner_points: - corner_centroid = centroid(corner_points) - corner_centroid.weight = settings.corner_points_weight / weight_pref_total - pois.append(corner_centroid) + corner_centroid = centroid(corner_points) + corner_centroid.weight = settings.corner_points_weight / weight_pref_total + pois.append(corner_centroid) entropy_centroid = None if entropy_points: - entropy_centroid = centroid(entropy_points) - entropy_centroid.weight = settings.entropy_points_weight / weight_pref_total - pois.append(entropy_centroid) + entropy_centroid = centroid(entropy_points) + entropy_centroid.weight = settings.entropy_points_weight / weight_pref_total + pois.append(entropy_centroid) face_centroid = None if face_points: - face_centroid = centroid(face_points) - face_centroid.weight = settings.face_points_weight / weight_pref_total - pois.append(face_centroid) + face_centroid = centroid(face_points) + face_centroid.weight = settings.face_points_weight / weight_pref_total + pois.append(face_centroid) average_point = poi_average(pois, settings) if settings.annotate_image: - d = ImageDraw.Draw(im) - max_size = min(im.width, im.height) * 0.07 - if corner_centroid is not None: - color = BLUE - box = corner_centroid.bounding(max_size * corner_centroid.weight) - d.text((box[0], box[1]-15), f"Edge: {corner_centroid.weight:.02f}", fill=color) - d.ellipse(box, outline=color) - if len(corner_points) > 1: - for f in corner_points: - d.rectangle(f.bounding(4), outline=color) - if entropy_centroid is not None: - color = "#ff0" - box = entropy_centroid.bounding(max_size * entropy_centroid.weight) - d.text((box[0], box[1]-15), f"Entropy: {entropy_centroid.weight:.02f}", fill=color) - d.ellipse(box, outline=color) - if len(entropy_points) > 1: - for f in entropy_points: - d.rectangle(f.bounding(4), outline=color) - if face_centroid is not None: - color = RED - box = face_centroid.bounding(max_size * face_centroid.weight) - d.text((box[0], box[1]-15), f"Face: {face_centroid.weight:.02f}", fill=color) - d.ellipse(box, outline=color) - if len(face_points) > 1: - for f in face_points: - d.rectangle(f.bounding(4), outline=color) + d = ImageDraw.Draw(im) + max_size = min(im.width, im.height) * 0.07 + if corner_centroid is not None: + color = BLUE + box = corner_centroid.bounding(max_size * corner_centroid.weight) + d.text((box[0], box[1] - 15), f"Edge: {corner_centroid.weight:.02f}", fill=color) + d.ellipse(box, outline=color) + if len(corner_points) > 1: + for f in corner_points: + d.rectangle(f.bounding(4), outline=color) + if entropy_centroid is not None: + color = "#ff0" + box = entropy_centroid.bounding(max_size * entropy_centroid.weight) + d.text((box[0], box[1] - 15), f"Entropy: {entropy_centroid.weight:.02f}", fill=color) + d.ellipse(box, outline=color) + if len(entropy_points) > 1: + for f in entropy_points: + d.rectangle(f.bounding(4), outline=color) + if face_centroid is not None: + color = RED + box = face_centroid.bounding(max_size * face_centroid.weight) + d.text((box[0], box[1] - 15), f"Face: {face_centroid.weight:.02f}", fill=color) + d.ellipse(box, outline=color) + if len(face_points) > 1: + for f in face_points: + d.rectangle(f.bounding(4), outline=color) - d.ellipse(average_point.bounding(max_size), outline=GREEN) + d.ellipse(average_point.bounding(max_size), outline=GREEN) return average_point def image_face_points(im, settings): if settings.dnn_model_path is not None: - detector = cv2.FaceDetectorYN.create( - settings.dnn_model_path, - "", - (im.width, im.height), - 0.9, # score threshold - 0.3, # nms threshold - 5000 # keep top k before nms - ) - faces = detector.detect(np.array(im)) - results = [] - if faces[1] is not None: - for face in faces[1]: - x = face[0] - y = face[1] - w = face[2] - h = face[3] - results.append( - PointOfInterest( - int(x + (w * 0.5)), # face focus left/right is center - int(y + (h * 0.33)), # face focus up/down is close to the top of the head - size = w, - weight = 1/len(faces[1]) - ) - ) - return results + detector = cv2.FaceDetectorYN.create( + settings.dnn_model_path, + "", + (im.width, im.height), + 0.9, # score threshold + 0.3, # nms threshold + 5000 # keep top k before nms + ) + faces = detector.detect(np.array(im)) + results = [] + if faces[1] is not None: + for face in faces[1]: + x = face[0] + y = face[1] + w = face[2] + h = face[3] + results.append( + PointOfInterest( + int(x + (w * 0.5)), # face focus left/right is center + int(y + (h * 0.33)), # face focus up/down is close to the top of the head + size=w, + weight=1 / len(faces[1]) + ) + ) + return results else: - np_im = np.array(im) - gray = cv2.cvtColor(np_im, cv2.COLOR_BGR2GRAY) + np_im = np.array(im) + gray = cv2.cvtColor(np_im, cv2.COLOR_BGR2GRAY) - tries = [ - [ f'{cv2.data.haarcascades}haarcascade_eye.xml', 0.01 ], - [ f'{cv2.data.haarcascades}haarcascade_frontalface_default.xml', 0.05 ], - [ f'{cv2.data.haarcascades}haarcascade_profileface.xml', 0.05 ], - [ f'{cv2.data.haarcascades}haarcascade_frontalface_alt.xml', 0.05 ], - [ f'{cv2.data.haarcascades}haarcascade_frontalface_alt2.xml', 0.05 ], - [ f'{cv2.data.haarcascades}haarcascade_frontalface_alt_tree.xml', 0.05 ], - [ f'{cv2.data.haarcascades}haarcascade_eye_tree_eyeglasses.xml', 0.05 ], - [ f'{cv2.data.haarcascades}haarcascade_upperbody.xml', 0.05 ] - ] - for t in tries: - classifier = cv2.CascadeClassifier(t[0]) - minsize = int(min(im.width, im.height) * t[1]) # at least N percent of the smallest side - try: - faces = classifier.detectMultiScale(gray, scaleFactor=1.1, - minNeighbors=7, minSize=(minsize, minsize), flags=cv2.CASCADE_SCALE_IMAGE) - except Exception: - continue + tries = [ + [f'{cv2.data.haarcascades}haarcascade_eye.xml', 0.01], + [f'{cv2.data.haarcascades}haarcascade_frontalface_default.xml', 0.05], + [f'{cv2.data.haarcascades}haarcascade_profileface.xml', 0.05], + [f'{cv2.data.haarcascades}haarcascade_frontalface_alt.xml', 0.05], + [f'{cv2.data.haarcascades}haarcascade_frontalface_alt2.xml', 0.05], + [f'{cv2.data.haarcascades}haarcascade_frontalface_alt_tree.xml', 0.05], + [f'{cv2.data.haarcascades}haarcascade_eye_tree_eyeglasses.xml', 0.05], + [f'{cv2.data.haarcascades}haarcascade_upperbody.xml', 0.05] + ] + for t in tries: + classifier = cv2.CascadeClassifier(t[0]) + minsize = int(min(im.width, im.height) * t[1]) # at least N percent of the smallest side + try: + faces = classifier.detectMultiScale(gray, scaleFactor=1.1, + minNeighbors=7, minSize=(minsize, minsize), + flags=cv2.CASCADE_SCALE_IMAGE) + except Exception: + continue - if faces: - rects = [[f[0], f[1], f[0] + f[2], f[1] + f[3]] for f in faces] - return [PointOfInterest((r[0] +r[2]) // 2, (r[1] + r[3]) // 2, size=abs(r[0]-r[2]), weight=1/len(rects)) for r in rects] + if faces: + rects = [[f[0], f[1], f[0] + f[2], f[1] + f[3]] for f in faces] + return [PointOfInterest((r[0] + r[2]) // 2, (r[1] + r[3]) // 2, size=abs(r[0] - r[2]), + weight=1 / len(rects)) for r in rects] return [] @@ -200,7 +202,7 @@ def image_corner_points(im, settings): # naive attempt at preventing focal points from collecting at watermarks near the bottom gd = ImageDraw.Draw(grayscale) - gd.rectangle([0, im.height*.9, im.width, im.height], fill="#999") + gd.rectangle([0, im.height * .9, im.width, im.height], fill="#999") np_im = np.array(grayscale) @@ -208,7 +210,7 @@ def image_corner_points(im, settings): np_im, maxCorners=100, qualityLevel=0.04, - minDistance=min(grayscale.width, grayscale.height)*0.06, + minDistance=min(grayscale.width, grayscale.height) * 0.06, useHarrisDetector=False, ) @@ -217,8 +219,8 @@ def image_corner_points(im, settings): focal_points = [] for point in points: - x, y = point.ravel() - focal_points.append(PointOfInterest(x, y, size=4, weight=1/len(points))) + x, y = point.ravel() + focal_points.append(PointOfInterest(x, y, size=4, weight=1 / len(points))) return focal_points @@ -227,13 +229,13 @@ def image_entropy_points(im, settings): landscape = im.height < im.width portrait = im.height > im.width if landscape: - move_idx = [0, 2] - move_max = im.size[0] + move_idx = [0, 2] + move_max = im.size[0] elif portrait: - move_idx = [1, 3] - move_max = im.size[1] + move_idx = [1, 3] + move_max = im.size[1] else: - return [] + return [] e_max = 0 crop_current = [0, 0, settings.crop_width, settings.crop_height] @@ -243,14 +245,14 @@ def image_entropy_points(im, settings): e = image_entropy(crop) if (e > e_max): - e_max = e - crop_best = list(crop_current) + e_max = e + crop_best = list(crop_current) crop_current[move_idx[0]] += 4 crop_current[move_idx[1]] += 4 - x_mid = int(crop_best[0] + settings.crop_width/2) - y_mid = int(crop_best[1] + settings.crop_height/2) + x_mid = int(crop_best[0] + settings.crop_width / 2) + y_mid = int(crop_best[1] + settings.crop_height / 2) return [PointOfInterest(x_mid, y_mid, size=25, weight=1.0)] From 39eae9f009c8302eed77b0942e1e634f6125d53e Mon Sep 17 00:00:00 2001 From: hidenorly Date: Wed, 29 Nov 2023 04:07:48 +0900 Subject: [PATCH 100/139] Revert "Add FP32 fallback support on sd_vae_approx" This reverts commit 58c19545c83fa6925c9ce2216ee64964eb5129ce. Since the modification is expected to move to mac_specific.py (https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14046#issuecomment-1826731532) --- modules/sd_vae_approx.py | 8 +------- 1 file changed, 1 insertion(+), 7 deletions(-) diff --git a/modules/sd_vae_approx.py b/modules/sd_vae_approx.py index 8370493f9..3965e223e 100644 --- a/modules/sd_vae_approx.py +++ b/modules/sd_vae_approx.py @@ -21,13 +21,7 @@ class VAEApprox(nn.Module): def forward(self, x): extra = 11 - try: - x = nn.functional.interpolate(x, (x.shape[2] * 2, x.shape[3] * 2)) - except RuntimeError as e: - if "not implemented for" in str(e) and "Half" in str(e): - x = nn.functional.interpolate(x.to(torch.float32), (x.shape[2] * 2, x.shape[3] * 2)).to(x.dtype) - else: - print(f"An unexpected RuntimeError occurred: {str(e)}") + x = nn.functional.interpolate(x, (x.shape[2] * 2, x.shape[3] * 2)) x = nn.functional.pad(x, (extra, extra, extra, extra)) for layer in [self.conv1, self.conv2, self.conv3, self.conv4, self.conv5, self.conv6, self.conv7, self.conv8, ]: From a0096c58977c01ddc6a2b83a8a7b64da6fd4a51e Mon Sep 17 00:00:00 2001 From: hidenorly Date: Wed, 29 Nov 2023 04:45:04 +0900 Subject: [PATCH 101/139] Add FP32 fallback support on torch.nn.functional.interpolate This tries to execute interpolate with FP32 if it failed. Background is that on some environment such as Mx chip MacOS devices, we get error as follows: ``` "torch/nn/functional.py", line 3931, in interpolate return torch._C._nn.upsample_nearest2d(input, output_size, scale_factors) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ RuntimeError: "upsample_nearest2d_channels_last" not implemented for 'Half' ``` In this case, ```--no-half``` doesn't help to solve. Therefore this commits add the FP32 fallback execution to solve it. Note that the ```upsample_nearest2d``` is called from ```torch.nn.functional.interpolate```. And the fallback for torch.nn.functional.interpolate is necessary at ```modules/sd_vae_approx.py``` 's ```VAEApprox.forward``` ```repositories/stable-diffusion-stability-ai/ldm/modules/diffusionmodules/openaimodel.py``` 's ```Upsample.forward``` --- modules/mac_specific.py | 16 ++++++++++++++++ 1 file changed, 16 insertions(+) diff --git a/modules/mac_specific.py b/modules/mac_specific.py index 89256c5b0..3538e659d 100644 --- a/modules/mac_specific.py +++ b/modules/mac_specific.py @@ -1,6 +1,8 @@ import logging import torch +from typing import Optional, List +from torch import Tensor import platform from modules.sd_hijack_utils import CondFunc from packaging import version @@ -51,6 +53,17 @@ def cumsum_fix(input, cumsum_func, *args, **kwargs): return cumsum_func(input, *args, **kwargs) +# MPS workaround for https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14046 +def interpolate_with_fp32_fallback(orig_func, *args, **kwargs) -> Tensor: + try: + return orig_func(*args, **kwargs) + except RuntimeError as e: + if "not implemented for" in str(e) and "Half" in str(e): + input_tensor = args[0] + return orig_func(input_tensor.to(torch.float32), *args[1:], **kwargs).to(input_tensor.dtype) + else: + print(f"An unexpected RuntimeError occurred: {str(e)}") + if has_mps: if platform.mac_ver()[0].startswith("13.2."): # MPS workaround for https://github.com/pytorch/pytorch/issues/95188, thanks to danieldk (https://github.com/explosion/curated-transformers/pull/124) @@ -77,6 +90,9 @@ if has_mps: # MPS workaround for https://github.com/pytorch/pytorch/issues/96113 CondFunc('torch.nn.functional.layer_norm', lambda orig_func, x, normalized_shape, weight, bias, eps, **kwargs: orig_func(x.float(), normalized_shape, weight.float() if weight is not None else None, bias.float() if bias is not None else bias, eps).to(x.dtype), lambda _, input, *args, **kwargs: len(args) == 4 and input.device.type == 'mps') + # MPS workaround for https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14046 + CondFunc('torch.nn.functional.interpolate', interpolate_with_fp32_fallback, None) + # MPS workaround for https://github.com/pytorch/pytorch/issues/92311 if platform.processor() == 'i386': for funcName in ['torch.argmax', 'torch.Tensor.argmax']: From 81c00728b8ec0b6c0e70ea10c7687aad065a95cb Mon Sep 17 00:00:00 2001 From: hidenorly Date: Wed, 29 Nov 2023 04:59:35 +0900 Subject: [PATCH 102/139] Fix the Ruff error about unused import --- modules/mac_specific.py | 1 - 1 file changed, 1 deletion(-) diff --git a/modules/mac_specific.py b/modules/mac_specific.py index 3538e659d..d96d86d79 100644 --- a/modules/mac_specific.py +++ b/modules/mac_specific.py @@ -1,7 +1,6 @@ import logging import torch -from typing import Optional, List from torch import Tensor import platform from modules.sd_hijack_utils import CondFunc From 8b40f475a31109cc6ecbdc0d14a0cee9e0303291 Mon Sep 17 00:00:00 2001 From: Nuullll Date: Fri, 10 Nov 2023 11:06:26 +0800 Subject: [PATCH 103/139] Initial IPEX support --- modules/devices.py | 11 +++++++++-- modules/xpu_specific.py | 42 +++++++++++++++++++++++++++++++++++++++++ 2 files changed, 51 insertions(+), 2 deletions(-) create mode 100644 modules/xpu_specific.py diff --git a/modules/devices.py b/modules/devices.py index 1d4eb5635..be599736c 100644 --- a/modules/devices.py +++ b/modules/devices.py @@ -3,7 +3,7 @@ import contextlib from functools import lru_cache import torch -from modules import errors, shared +from modules import errors, shared, xpu_specific if sys.platform == "darwin": from modules import mac_specific @@ -30,6 +30,9 @@ def get_optimal_device_name(): if has_mps(): return "mps" + if xpu_specific.has_ipex: + return xpu_specific.get_xpu_device_string() + return "cpu" @@ -100,11 +103,15 @@ def autocast(disable=False): if dtype == torch.float32 or shared.cmd_opts.precision == "full": return contextlib.nullcontext() + if xpu_specific.has_xpu: + return torch.autocast("xpu") + return torch.autocast("cuda") def without_autocast(disable=False): - return torch.autocast("cuda", enabled=False) if torch.is_autocast_enabled() and not disable else contextlib.nullcontext() + device_type = "xpu" if xpu_specific.has_xpu else "cuda" + return torch.autocast(device_type, enabled=False) if torch.is_autocast_enabled() and not disable else contextlib.nullcontext() class NansException(Exception): diff --git a/modules/xpu_specific.py b/modules/xpu_specific.py new file mode 100644 index 000000000..6417dd2d6 --- /dev/null +++ b/modules/xpu_specific.py @@ -0,0 +1,42 @@ +import contextlib +from modules import shared +from modules.sd_hijack_utils import CondFunc + +has_ipex = False +try: + import torch + import intel_extension_for_pytorch as ipex + has_ipex = True +except Exception: + pass + +def check_for_xpu(): + if not has_ipex: + return False + + return hasattr(torch, 'xpu') and torch.xpu.is_available() + +has_xpu = check_for_xpu() + +def get_xpu_device_string(): + if shared.cmd_opts.device_id is not None: + return f"xpu:{shared.cmd_opts.device_id}" + return "xpu" + +def return_null_context(*args, **kwargs): # pylint: disable=unused-argument + return contextlib.nullcontext() + +if has_xpu: + CondFunc('torch.Generator', + lambda orig_func, device=None: torch.xpu.Generator(device), + lambda orig_func, device=None: device is not None and device != torch.device("cpu") and device != "cpu") + + CondFunc('torch.nn.functional.layer_norm', + lambda orig_func, input, normalized_shape=None, weight=None, *args, **kwargs: + orig_func(input.to(weight.data.dtype), normalized_shape, weight, *args, **kwargs), + lambda orig_func, input, normalized_shape=None, weight=None, *args, **kwargs: + weight is not None and input.dtype != weight.data.dtype) + + CondFunc('torch.nn.modules.GroupNorm.forward', + lambda orig_func, self, input: orig_func(self, input.to(self.weight.data.dtype)), + lambda orig_func, self, input: input.dtype != self.weight.data.dtype) From c2ed4132037a32cda856e8ba6e2cda32b44b9784 Mon Sep 17 00:00:00 2001 From: w-e-w <40751091+w-e-w@users.noreply.github.com> Date: Fri, 1 Dec 2023 02:59:41 +0900 Subject: [PATCH 104/139] add max-heigh/width to global-popup-inner prevent the pop-up from being too big as to making exiting the pop-up impossible --- style.css | 2 ++ 1 file changed, 2 insertions(+) diff --git a/style.css b/style.css index 6e3ca8411..ee39a57b7 100644 --- a/style.css +++ b/style.css @@ -646,6 +646,8 @@ table.popup-table .link{ margin: auto; padding: 2em; z-index: 1001; + max-height: 90%; + max-width: 90%; } /* fullpage image viewer */ From 01c8f1803a77c63b2ebfd3cbbd41659fb914f274 Mon Sep 17 00:00:00 2001 From: missionfloyd Date: Thu, 30 Nov 2023 22:36:12 -0700 Subject: [PATCH 105/139] Close popups with escape key --- javascript/extraNetworks.js | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/javascript/extraNetworks.js b/javascript/extraNetworks.js index a787372cf..98a7abb74 100644 --- a/javascript/extraNetworks.js +++ b/javascript/extraNetworks.js @@ -392,3 +392,9 @@ function extraNetworksRefreshSingleCard(page, tabname, name) { } }); } + +window.addEventListener("keydown", function(event) { + if (event.key == "Escape") { + closePopup(); + } +}); From 293f44e6c1de7bbf744a4236db81ac4559bdb82a Mon Sep 17 00:00:00 2001 From: MrCheeze Date: Fri, 1 Dec 2023 22:56:08 -0500 Subject: [PATCH 106/139] Fix bug where is_using_v_parameterization_for_sd2 fails because the sd_hijack is only partially undone --- modules/sd_hijack.py | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-) diff --git a/modules/sd_hijack.py b/modules/sd_hijack.py index 0157e19f0..3d340fc9b 100644 --- a/modules/sd_hijack.py +++ b/modules/sd_hijack.py @@ -38,9 +38,6 @@ ldm.models.diffusion.ddpm.print = shared.ldm_print optimizers = [] current_optimizer: sd_hijack_optimizations.SdOptimization = None -ldm_original_forward = patches.patch(__file__, ldm.modules.diffusionmodules.openaimodel.UNetModel, "forward", sd_unet.UNetModel_forward) -sgm_original_forward = patches.patch(__file__, sgm.modules.diffusionmodules.openaimodel.UNetModel, "forward", sd_unet.UNetModel_forward) - def list_optimizers(): new_optimizers = script_callbacks.list_optimizers_callback() @@ -258,6 +255,9 @@ class StableDiffusionModelHijack: import modules.models.diffusion.ddpm_edit + ldm_original_forward = patches.patch(__file__, ldm.modules.diffusionmodules.openaimodel.UNetModel, "forward", sd_unet.UNetModel_forward) + sgm_original_forward = patches.patch(__file__, sgm.modules.diffusionmodules.openaimodel.UNetModel, "forward", sd_unet.UNetModel_forward) + if isinstance(m, ldm.models.diffusion.ddpm.LatentDiffusion): sd_unet.original_forward = ldm_original_forward elif isinstance(m, modules.models.diffusion.ddpm_edit.LatentDiffusion): @@ -303,6 +303,9 @@ class StableDiffusionModelHijack: self.layers = None self.clip = None + patches.undo(__file__, ldm.modules.diffusionmodules.openaimodel.UNetModel, "forward") + patches.undo(__file__, sgm.modules.diffusionmodules.openaimodel.UNetModel, "forward") + sd_unet.original_forward = None From 6080045b2a0964e63bdcd33dd26015f8a51411f6 Mon Sep 17 00:00:00 2001 From: MrCheeze Date: Fri, 1 Dec 2023 22:58:05 -0500 Subject: [PATCH 107/139] Add support for SD 2.1 Turbo, by converting the state dict from SGM to LDM on load --- modules/sd_models.py | 17 +++++++++++++---- 1 file changed, 13 insertions(+), 4 deletions(-) diff --git a/modules/sd_models.py b/modules/sd_models.py index 841402e86..9355f1e16 100644 --- a/modules/sd_models.py +++ b/modules/sd_models.py @@ -230,15 +230,19 @@ def select_checkpoint(): return checkpoint_info -checkpoint_dict_replacements = { +checkpoint_dict_replacements_sd1 = { 'cond_stage_model.transformer.embeddings.': 'cond_stage_model.transformer.text_model.embeddings.', 'cond_stage_model.transformer.encoder.': 'cond_stage_model.transformer.text_model.encoder.', 'cond_stage_model.transformer.final_layer_norm.': 'cond_stage_model.transformer.text_model.final_layer_norm.', } +checkpoint_dict_replacements_sd2_turbo = { # Converts SD 2.1 Turbo from SGM to LDM format. + 'conditioner.embedders.0.': 'cond_stage_model.', +} -def transform_checkpoint_dict_key(k): - for text, replacement in checkpoint_dict_replacements.items(): + +def transform_checkpoint_dict_key(k, replacements): + for text, replacement in replacements.items(): if k.startswith(text): k = replacement + k[len(text):] @@ -249,9 +253,14 @@ def get_state_dict_from_checkpoint(pl_sd): pl_sd = pl_sd.pop("state_dict", pl_sd) pl_sd.pop("state_dict", None) + is_sd2_turbo = 'conditioner.embedders.0.model.ln_final.weight' in pl_sd and pl_sd['conditioner.embedders.0.model.ln_final.weight'].size()[0] == 1024 + sd = {} for k, v in pl_sd.items(): - new_key = transform_checkpoint_dict_key(k) + if is_sd2_turbo: + new_key = transform_checkpoint_dict_key(k, checkpoint_dict_replacements_sd2_turbo) + else: + new_key = transform_checkpoint_dict_key(k, checkpoint_dict_replacements_sd1) if new_key is not None: sd[new_key] = v From b58d061e41cba6fb91910d310d53e175d0511650 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Sat, 2 Dec 2023 08:33:28 +0300 Subject: [PATCH 108/139] infotext updates: add option to disregard certain infotext fields, add option to not include VAE in infotext, add explanation to infotext settings page, move some options to infotext settings page --- modules/generation_parameters_copypaste.py | 13 +++++++++---- modules/processing.py | 4 ++-- modules/shared_items.py | 16 ++++++++++++++++ modules/shared_options.py | 20 ++++++++++++++------ 4 files changed, 41 insertions(+), 12 deletions(-) diff --git a/modules/generation_parameters_copypaste.py b/modules/generation_parameters_copypaste.py index 0a606515b..4efe53e0c 100644 --- a/modules/generation_parameters_copypaste.py +++ b/modules/generation_parameters_copypaste.py @@ -1,3 +1,4 @@ +from __future__ import annotations import base64 import io import json @@ -15,9 +16,6 @@ re_imagesize = re.compile(r"^(\d+)x(\d+)$") re_hypernet_hash = re.compile("\(([0-9a-f]+)\)$") type_of_gr_update = type(gr.update()) -paste_fields = {} -registered_param_bindings = [] - class ParamBinding: def __init__(self, paste_button, tabname, source_text_component=None, source_image_component=None, source_tabname=None, override_settings_component=None, paste_field_names=None): @@ -30,6 +28,10 @@ class ParamBinding: self.paste_field_names = paste_field_names or [] +paste_fields: dict[str, dict] = {} +registered_param_bindings: list[ParamBinding] = [] + + def reset(): paste_fields.clear() registered_param_bindings.clear() @@ -113,7 +115,6 @@ def register_paste_params_button(binding: ParamBinding): def connect_paste_params_buttons(): - binding: ParamBinding for binding in registered_param_bindings: destination_image_component = paste_fields[binding.tabname]["init_img"] fields = paste_fields[binding.tabname]["fields"] @@ -313,6 +314,9 @@ Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 965400086, Size: 512x512, Model if "VAE Decoder" not in res: res["VAE Decoder"] = "Full" + skip = set(shared.opts.infotext_skip_pasting) + res = {k: v for k, v in res.items() if k not in skip} + return res @@ -443,3 +447,4 @@ def connect_paste(button, paste_fields, input_comp, override_settings_component, outputs=[], show_progress=False, ) + diff --git a/modules/processing.py b/modules/processing.py index ac58ef869..5ab6dddef 100644 --- a/modules/processing.py +++ b/modules/processing.py @@ -679,8 +679,8 @@ def create_infotext(p, all_prompts, all_seeds, all_subseeds, comments=None, iter "Size": f"{p.width}x{p.height}", "Model hash": p.sd_model_hash if opts.add_model_hash_to_info else None, "Model": p.sd_model_name if opts.add_model_name_to_info else None, - "VAE hash": p.sd_vae_hash if opts.add_model_hash_to_info else None, - "VAE": p.sd_vae_name if opts.add_model_name_to_info else None, + "VAE hash": p.sd_vae_hash if opts.add_vae_hash_to_info else None, + "VAE": p.sd_vae_name if opts.add_vae_name_to_info else None, "Variation seed": (None if p.subseed_strength == 0 else (p.all_subseeds[0] if use_main_prompt else all_subseeds[index])), "Variation seed strength": (None if p.subseed_strength == 0 else p.subseed_strength), "Seed resize from": (None if p.seed_resize_from_w <= 0 or p.seed_resize_from_h <= 0 else f"{p.seed_resize_from_w}x{p.seed_resize_from_h}"), diff --git a/modules/shared_items.py b/modules/shared_items.py index 5024b4268..991971ad0 100644 --- a/modules/shared_items.py +++ b/modules/shared_items.py @@ -66,6 +66,22 @@ def reload_hypernetworks(): shared.hypernetworks = hypernetwork.list_hypernetworks(cmd_opts.hypernetwork_dir) +def get_infotext_names(): + from modules import generation_parameters_copypaste, shared + res = {} + + for info in shared.opts.data_labels.values(): + if info.infotext: + res[info.infotext] = 1 + + for tab_data in generation_parameters_copypaste.paste_fields.values(): + for _, name in tab_data.get("fields") or []: + if isinstance(name, str): + res[name] = 1 + + return list(res) + + ui_reorder_categories_builtin_items = [ "prompt", "image", diff --git a/modules/shared_options.py b/modules/shared_options.py index 04e68a712..df45fc0a0 100644 --- a/modules/shared_options.py +++ b/modules/shared_options.py @@ -46,8 +46,6 @@ options_templates.update(options_section(('saving-images', "Saving images/grids" "grid_text_inactive_color": OptionInfo("#999999", "Inactive text color for image grids", ui_components.FormColorPicker, {}), "grid_background_color": OptionInfo("#ffffff", "Background color for image grids", ui_components.FormColorPicker, {}), - "enable_pnginfo": OptionInfo(True, "Save text information about generation parameters as chunks to png files"), - "save_txt": OptionInfo(False, "Create a text file next to every image with generation parameters."), "save_images_before_face_restoration": OptionInfo(False, "Save a copy of image before doing face restoration."), "save_images_before_highres_fix": OptionInfo(False, "Save a copy of image before applying highres fix."), "save_images_before_color_correction": OptionInfo(False, "Save a copy of image before applying color correction to img2img results"), @@ -288,11 +286,21 @@ options_templates.update(options_section(('ui', "User interface", "ui"), { options_templates.update(options_section(('infotext', "Infotext", "ui"), { - "add_model_hash_to_info": OptionInfo(True, "Add model hash to generation information"), - "add_model_name_to_info": OptionInfo(True, "Add model name to generation information"), - "add_user_name_to_info": OptionInfo(False, "Add user name to generation information when authenticated"), - "add_version_to_infotext": OptionInfo(True, "Add program version to generation information"), + "infotext_explanation": OptionHTML(""" +Infotext is what this software calls the text that contains generation parameters and can be used to generate the same picture again. +It is displayed in UI below the image. To use infotext, paste it into the prompt and click the ↙️ paste button. +"""), + "enable_pnginfo": OptionInfo(True, "Write infotext to metadata of the generated image"), + "save_txt": OptionInfo(False, "Create a text file with infotext next to every generated image"), + + "add_model_name_to_info": OptionInfo(True, "Add model name to infotext"), + "add_model_hash_to_info": OptionInfo(True, "Add model hash to infotext"), + "add_vae_name_to_info": OptionInfo(True, "Add VAE name to infotext"), + "add_vae_hash_to_info": OptionInfo(True, "Add VAE hash to infotext"), + "add_user_name_to_info": OptionInfo(False, "Add user name to infotext when authenticated"), + "add_version_to_infotext": OptionInfo(True, "Add program version to infotext"), "disable_weights_auto_swap": OptionInfo(True, "Disregard checkpoint information from pasted infotext").info("when reading generation parameters from text into UI"), + "infotext_skip_pasting": OptionInfo([], "Disregard fields from pasted infotext", ui_components.DropdownMulti, lambda: {"choices": shared_items.get_infotext_names()}), "infotext_styles": OptionInfo("Apply if any", "Infer styles from prompts of pasted infotext", gr.Radio, {"choices": ["Ignore", "Apply", "Discard", "Apply if any"]}).info("when reading generation parameters from text into UI)").html("""
  • Ignore: keep prompt and styles dropdown as it is.
  • Apply: remove style text from prompt, always replace styles dropdown value with found styles (even if none are found).
    • From 7499148ad4dbd3444215c843d02453f68c459707 Mon Sep 17 00:00:00 2001 From: Nuullll Date: Sat, 2 Dec 2023 14:00:46 +0800 Subject: [PATCH 109/139] Disable ipex autocast due to its bad perf --- modules/cmd_args.py | 1 + modules/devices.py | 20 +++++++++++++------- modules/xpu_specific.py | 28 ++++++++++++++++++---------- webui-ipex-user.bat | 19 +++++++++++++++++++ 4 files changed, 51 insertions(+), 17 deletions(-) create mode 100644 webui-ipex-user.bat diff --git a/modules/cmd_args.py b/modules/cmd_args.py index a9fb9bfa3..da93eb266 100644 --- a/modules/cmd_args.py +++ b/modules/cmd_args.py @@ -70,6 +70,7 @@ parser.add_argument("--opt-sdp-no-mem-attention", action='store_true', help="pre parser.add_argument("--disable-opt-split-attention", action='store_true', help="prefer no cross-attention layer optimization for automatic choice of optimization") parser.add_argument("--disable-nan-check", action='store_true', help="do not check if produced images/latent spaces have nans; useful for running without a checkpoint in CI") parser.add_argument("--use-cpu", nargs='+', help="use CPU as torch device for specified modules", default=[], type=str.lower) +parser.add_argument("--use-ipex", action="store_true", help="use Intel XPU as torch device") parser.add_argument("--disable-model-loading-ram-optimization", action='store_true', help="disable an optimization that reduces RAM use when loading a model") parser.add_argument("--listen", action='store_true', help="launch gradio with 0.0.0.0 as server name, allowing to respond to network requests") parser.add_argument("--port", type=int, help="launch gradio with given server port, you need root/admin rights for ports < 1024, defaults to 7860 if available", default=None) diff --git a/modules/devices.py b/modules/devices.py index be599736c..37ecca784 100644 --- a/modules/devices.py +++ b/modules/devices.py @@ -3,11 +3,18 @@ import contextlib from functools import lru_cache import torch -from modules import errors, shared, xpu_specific +from modules import errors, shared if sys.platform == "darwin": from modules import mac_specific +if shared.cmd_opts.use_ipex: + from modules import xpu_specific + + +def has_xpu() -> bool: + return shared.cmd_opts.use_ipex and xpu_specific.has_xpu + def has_mps() -> bool: if sys.platform != "darwin": @@ -30,7 +37,7 @@ def get_optimal_device_name(): if has_mps(): return "mps" - if xpu_specific.has_ipex: + if has_xpu(): return xpu_specific.get_xpu_device_string() return "cpu" @@ -57,6 +64,9 @@ def torch_gc(): if has_mps(): mac_specific.torch_mps_gc() + if has_xpu(): + xpu_specific.torch_xpu_gc() + def enable_tf32(): if torch.cuda.is_available(): @@ -103,15 +113,11 @@ def autocast(disable=False): if dtype == torch.float32 or shared.cmd_opts.precision == "full": return contextlib.nullcontext() - if xpu_specific.has_xpu: - return torch.autocast("xpu") - return torch.autocast("cuda") def without_autocast(disable=False): - device_type = "xpu" if xpu_specific.has_xpu else "cuda" - return torch.autocast(device_type, enabled=False) if torch.is_autocast_enabled() and not disable else contextlib.nullcontext() + return torch.autocast("cuda", enabled=False) if torch.is_autocast_enabled() and not disable else contextlib.nullcontext() class NansException(Exception): diff --git a/modules/xpu_specific.py b/modules/xpu_specific.py index 6417dd2d6..2df68665a 100644 --- a/modules/xpu_specific.py +++ b/modules/xpu_specific.py @@ -1,4 +1,3 @@ -import contextlib from modules import shared from modules.sd_hijack_utils import CondFunc @@ -10,33 +9,42 @@ try: except Exception: pass + def check_for_xpu(): - if not has_ipex: - return False + return has_ipex and hasattr(torch, 'xpu') and torch.xpu.is_available() - return hasattr(torch, 'xpu') and torch.xpu.is_available() - -has_xpu = check_for_xpu() def get_xpu_device_string(): if shared.cmd_opts.device_id is not None: return f"xpu:{shared.cmd_opts.device_id}" return "xpu" -def return_null_context(*args, **kwargs): # pylint: disable=unused-argument - return contextlib.nullcontext() + +def torch_xpu_gc(): + with torch.xpu.device(get_xpu_device_string()): + torch.xpu.empty_cache() + + +has_xpu = check_for_xpu() if has_xpu: + # W/A for https://github.com/intel/intel-extension-for-pytorch/issues/452: torch.Generator API doesn't support XPU device CondFunc('torch.Generator', lambda orig_func, device=None: torch.xpu.Generator(device), - lambda orig_func, device=None: device is not None and device != torch.device("cpu") and device != "cpu") + lambda orig_func, device=None: device is not None and device.type == "xpu") + # W/A for some OPs that could not handle different input dtypes CondFunc('torch.nn.functional.layer_norm', lambda orig_func, input, normalized_shape=None, weight=None, *args, **kwargs: orig_func(input.to(weight.data.dtype), normalized_shape, weight, *args, **kwargs), lambda orig_func, input, normalized_shape=None, weight=None, *args, **kwargs: weight is not None and input.dtype != weight.data.dtype) - CondFunc('torch.nn.modules.GroupNorm.forward', lambda orig_func, self, input: orig_func(self, input.to(self.weight.data.dtype)), lambda orig_func, self, input: input.dtype != self.weight.data.dtype) + CondFunc('torch.nn.modules.linear.Linear.forward', + lambda orig_func, self, input: orig_func(self, input.to(self.weight.data.dtype)), + lambda orig_func, self, input: input.dtype != self.weight.data.dtype) + CondFunc('torch.nn.modules.conv.Conv2d.forward', + lambda orig_func, self, input: orig_func(self, input.to(self.weight.data.dtype)), + lambda orig_func, self, input: input.dtype != self.weight.data.dtype) diff --git a/webui-ipex-user.bat b/webui-ipex-user.bat new file mode 100644 index 000000000..ab25a0400 --- /dev/null +++ b/webui-ipex-user.bat @@ -0,0 +1,19 @@ +@echo off + +set PYTHON= +@REM The "Nuullll/intel-extension-for-pytorch" wheels were built from IPEX source for Intel Arc GPU: https://github.com/intel/intel-extension-for-pytorch/tree/xpu-main +@REM This is NOT an Intel official release so please use it at your own risk!! +@REM See https://github.com/Nuullll/intel-extension-for-pytorch/releases/tag/v2.0.110%2Bxpu-master%2Bdll-bundle for details. +@REM +@REM Strengths (over official IPEX 2.0.110 windows release): +@REM - AOT build (for Arc GPU only) to eliminate JIT compilation overhead: https://github.com/intel/intel-extension-for-pytorch/issues/399 +@REM - Bundles minimal oneAPI 2023.2 dependencies into the python wheels, so users don't need to install oneAPI for the whole system. +@REM - Provides a compatible torchvision wheel: https://github.com/intel/intel-extension-for-pytorch/issues/465 +@REM Limitation: +@REM - Only works for python 3.10 +set "TORCH_COMMAND=pip install https://github.com/Nuullll/intel-extension-for-pytorch/releases/download/v2.0.110%%2Bxpu-master%%2Bdll-bundle/torch-2.0.0a0+gite9ebda2-cp310-cp310-win_amd64.whl https://github.com/Nuullll/intel-extension-for-pytorch/releases/download/v2.0.110%%2Bxpu-master%%2Bdll-bundle/torchvision-0.15.2a0+fa99a53-cp310-cp310-win_amd64.whl https://github.com/Nuullll/intel-extension-for-pytorch/releases/download/v2.0.110%%2Bxpu-master%%2Bdll-bundle/intel_extension_for_pytorch-2.0.110+gitc6ea20b-cp310-cp310-win_amd64.whl" +set GIT= +set VENV_DIR= +set "COMMANDLINE_ARGS=--use-ipex --skip-torch-cuda-test --skip-version-check --opt-sdp-attention" + +call webui.bat From e294e46d46a814457fc77af13c17128bd6075d45 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Sat, 2 Dec 2023 09:26:38 +0300 Subject: [PATCH 110/139] split UI settings page into many --- .../scripts/extra_options_section.py | 13 +++-- modules/shared_options.py | 57 +++++++++++-------- 2 files changed, 40 insertions(+), 30 deletions(-) diff --git a/extensions-builtin/extra-options-section/scripts/extra_options_section.py b/extensions-builtin/extra-options-section/scripts/extra_options_section.py index 983f87ff0..a903df625 100644 --- a/extensions-builtin/extra-options-section/scripts/extra_options_section.py +++ b/extensions-builtin/extra-options-section/scripts/extra_options_section.py @@ -64,11 +64,14 @@ class ExtraOptionsSection(scripts.Script): p.override_settings[name] = value -shared.options_templates.update(shared.options_section(('ui', "User interface"), { - "extra_options_txt2img": shared.OptionInfo([], "Options in main UI - txt2img", ui_components.DropdownMulti, lambda: {"choices": list(shared.opts.data_labels.keys())}).js("info", "settingsHintsShowQuicksettings").info("setting entries that also appear in txt2img interfaces").needs_reload_ui(), - "extra_options_img2img": shared.OptionInfo([], "Options in main UI - img2img", ui_components.DropdownMulti, lambda: {"choices": list(shared.opts.data_labels.keys())}).js("info", "settingsHintsShowQuicksettings").info("setting entries that also appear in img2img interfaces").needs_reload_ui(), - "extra_options_cols": shared.OptionInfo(1, "Options in main UI - number of columns", gr.Number, {"precision": 0}).needs_reload_ui(), - "extra_options_accordion": shared.OptionInfo(False, "Options in main UI - place into an accordion").needs_reload_ui() +shared.options_templates.update(shared.options_section(('settings_in_ui', "Settings in UI", "ui"), { + "settings_in_ui": shared.OptionHTML(""" +This page allows you to add some settings to the main interface of txt2img and img2img tabs. +"""), + "extra_options_txt2img": shared.OptionInfo([], "Settings for txt2img", ui_components.DropdownMulti, lambda: {"choices": list(shared.opts.data_labels.keys())}).js("info", "settingsHintsShowQuicksettings").info("setting entries that also appear in txt2img interfaces").needs_reload_ui(), + "extra_options_img2img": shared.OptionInfo([], "Settings for img2img", ui_components.DropdownMulti, lambda: {"choices": list(shared.opts.data_labels.keys())}).js("info", "settingsHintsShowQuicksettings").info("setting entries that also appear in img2img interfaces").needs_reload_ui(), + "extra_options_cols": shared.OptionInfo(1, "Number of columns for added settings", gr.Number, {"precision": 0}).needs_reload_ui(), + "extra_options_accordion": shared.OptionInfo(False, "Place added settings into an accordion").needs_reload_ui() })) diff --git a/modules/shared_options.py b/modules/shared_options.py index df45fc0a0..1390152db 100644 --- a/modules/shared_options.py +++ b/modules/shared_options.py @@ -250,38 +250,45 @@ options_templates.update(options_section(('extra_networks', "Extra Networks", "s "sd_hypernetwork": OptionInfo("None", "Add hypernetwork to prompt", gr.Dropdown, lambda: {"choices": ["None", *shared.hypernetworks]}, refresh=shared_items.reload_hypernetworks), })) -options_templates.update(options_section(('ui', "User interface", "ui"), { - "localization": OptionInfo("None", "Localization", gr.Dropdown, lambda: {"choices": ["None"] + list(localization.localizations.keys())}, refresh=lambda: localization.list_localizations(cmd_opts.localizations_dir)).needs_reload_ui(), - "gradio_theme": OptionInfo("Default", "Gradio theme", ui_components.DropdownEditable, lambda: {"choices": ["Default"] + shared_gradio_themes.gradio_hf_hub_themes}).info("you can also manually enter any of themes from the gallery.").needs_reload_ui(), - "gradio_themes_cache": OptionInfo(True, "Cache gradio themes locally").info("disable to update the selected Gradio theme"), - "gallery_height": OptionInfo("", "Gallery height", gr.Textbox).info("an be any valid CSS value").needs_reload_ui(), - "return_grid": OptionInfo(True, "Show grid in results for web"), - "do_not_show_images": OptionInfo(False, "Do not show any images in results for web"), - "send_seed": OptionInfo(True, "Send seed when sending prompt or image to other interface"), - "send_size": OptionInfo(True, "Send size when sending prompt or image to another interface"), - "js_modal_lightbox": OptionInfo(True, "Enable full page image viewer"), - "js_modal_lightbox_initially_zoomed": OptionInfo(True, "Show images zoomed in by default in full page image viewer"), - "js_modal_lightbox_gamepad": OptionInfo(False, "Navigate image viewer with gamepad"), - "js_modal_lightbox_gamepad_repeat": OptionInfo(250, "Gamepad repeat period, in milliseconds"), - "show_progress_in_title": OptionInfo(True, "Show generation progress in window title."), +options_templates.update(options_section(('ui_prompt_editing', "Prompt editing", "ui"), { + "keyedit_precision_attention": OptionInfo(0.1, "Precision for (attention:1.1) when editing the prompt with Ctrl+up/down", gr.Slider, {"minimum": 0.01, "maximum": 0.2, "step": 0.001}), + "keyedit_precision_extra": OptionInfo(0.05, "Precision for when editing the prompt with Ctrl+up/down", gr.Slider, {"minimum": 0.01, "maximum": 0.2, "step": 0.001}), + "keyedit_delimiters": OptionInfo(r".,\/!?%^*;:{}=`~() ", "Word delimiters when editing the prompt with Ctrl+up/down"), + "disable_token_counters": OptionInfo(False, "Disable prompt token counters").needs_reload_ui(), +})) + +options_templates.update(options_section(('ui_gallery', "Gallery", "ui"), { + "return_grid": OptionInfo(True, "Show grid in gallery"), + "do_not_show_images": OptionInfo(False, "Do not show any images in gallery"), + "js_modal_lightbox": OptionInfo(True, "Full page image viewer: enable"), + "js_modal_lightbox_initially_zoomed": OptionInfo(True, "Full page image viewer: show images zoomed in by default"), + "js_modal_lightbox_gamepad": OptionInfo(False, "Full page image viewer: navigate with gamepad"), + "js_modal_lightbox_gamepad_repeat": OptionInfo(250, "Full page image viewer: gamepad repeat period").info("in milliseconds"), + "gallery_height": OptionInfo("", "Gallery height", gr.Textbox).info("can be any valid CSS value, for example 768px or 20em").needs_reload_ui(), +})) + +options_templates.update(options_section(('ui_alternatives', "UI alternatives", "ui"), { + "compact_prompt_box": OptionInfo(False, "Compact prompt layout").info("puts prompt and negative prompt inside the Generate tab, leaving more vertical space for the image on the right").needs_reload_ui(), "samplers_in_dropdown": OptionInfo(True, "Use dropdown for sampler selection instead of radio group").needs_reload_ui(), "dimensions_and_batch_together": OptionInfo(True, "Show Width/Height and Batch sliders in same row").needs_reload_ui(), - "keyedit_precision_attention": OptionInfo(0.1, "Ctrl+up/down precision when editing (attention:1.1)", gr.Slider, {"minimum": 0.01, "maximum": 0.2, "step": 0.001}), - "keyedit_precision_extra": OptionInfo(0.05, "Ctrl+up/down precision when editing ", gr.Slider, {"minimum": 0.01, "maximum": 0.2, "step": 0.001}), - "keyedit_delimiters": OptionInfo(r".,\/!?%^*;:{}=`~() ", "Ctrl+up/down word delimiters"), - "keyedit_delimiters_whitespace": OptionInfo(["Tab", "Carriage Return", "Line Feed"], "Ctrl+up/down whitespace delimiters", gr.CheckboxGroup, lambda: {"choices": ["Tab", "Carriage Return", "Line Feed"]}), - "keyedit_move": OptionInfo(True, "Alt+left/right moves prompt elements"), - "quicksettings_list": OptionInfo(["sd_model_checkpoint"], "Quicksettings list", ui_components.DropdownMulti, lambda: {"choices": list(shared.opts.data_labels.keys())}).js("info", "settingsHintsShowQuicksettings").info("setting entries that appear at the top of page rather than in settings tab").needs_reload_ui(), - "ui_tab_order": OptionInfo([], "UI tab order", ui_components.DropdownMulti, lambda: {"choices": list(shared.tab_names)}).needs_reload_ui(), - "hidden_tabs": OptionInfo([], "Hidden UI tabs", ui_components.DropdownMulti, lambda: {"choices": list(shared.tab_names)}).needs_reload_ui(), - "ui_reorder_list": OptionInfo([], "txt2img/img2img UI item order", ui_components.DropdownMulti, lambda: {"choices": list(shared_items.ui_reorder_categories())}).info("selected items appear first").needs_reload_ui(), "sd_checkpoint_dropdown_use_short": OptionInfo(False, "Checkpoint dropdown: use filenames without paths").info("models in subdirectories like photo/sd15.ckpt will be listed as just sd15.ckpt"), "hires_fix_show_sampler": OptionInfo(False, "Hires fix: show hires checkpoint and sampler selection").needs_reload_ui(), "hires_fix_show_prompts": OptionInfo(False, "Hires fix: show hires prompt and negative prompt").needs_reload_ui(), - "disable_token_counters": OptionInfo(False, "Disable prompt token counters").needs_reload_ui(), "txt2img_settings_accordion": OptionInfo(False, "Settings in txt2img hidden under Accordion").needs_reload_ui(), "img2img_settings_accordion": OptionInfo(False, "Settings in img2img hidden under Accordion").needs_reload_ui(), - "compact_prompt_box": OptionInfo(False, "Compact prompt layout").info("puts prompt and negative prompt inside the Generate tab, leaving more vertical space for the image on the right").needs_reload_ui(), +})) + +options_templates.update(options_section(('ui', "User interface", "ui"), { + "localization": OptionInfo("None", "Localization", gr.Dropdown, lambda: {"choices": ["None"] + list(localization.localizations.keys())}, refresh=lambda: localization.list_localizations(cmd_opts.localizations_dir)).needs_reload_ui(), + "quicksettings_list": OptionInfo(["sd_model_checkpoint"], "Quicksettings list", ui_components.DropdownMulti, lambda: {"choices": list(shared.opts.data_labels.keys())}).js("info", "settingsHintsShowQuicksettings").info("setting entries that appear at the top of page rather than in settings tab").needs_reload_ui(), + "ui_tab_order": OptionInfo([], "UI tab order", ui_components.DropdownMulti, lambda: {"choices": list(shared.tab_names)}).needs_reload_ui(), + "hidden_tabs": OptionInfo([], "Hidden UI tabs", ui_components.DropdownMulti, lambda: {"choices": list(shared.tab_names)}).needs_reload_ui(), + "ui_reorder_list": OptionInfo([], "UI item order for txt2img/img2img tabs", ui_components.DropdownMulti, lambda: {"choices": list(shared_items.ui_reorder_categories())}).info("selected items appear first").needs_reload_ui(), + "gradio_theme": OptionInfo("Default", "Gradio theme", ui_components.DropdownEditable, lambda: {"choices": ["Default"] + shared_gradio_themes.gradio_hf_hub_themes}).info("you can also manually enter any of themes from the gallery.").needs_reload_ui(), + "gradio_themes_cache": OptionInfo(True, "Cache gradio themes locally").info("disable to update the selected Gradio theme"), + "show_progress_in_title": OptionInfo(True, "Show generation progress in window title."), + "send_seed": OptionInfo(True, "Send seed when sending prompt or image to other interface"), + "send_size": OptionInfo(True, "Send size when sending prompt or image to another interface"), })) From ef6b8123dc57e4e4bd5e08d9f3e3dbdfdf6b4c4a Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Sat, 2 Dec 2023 09:57:39 +0300 Subject: [PATCH 111/139] put code that can cause an exception into its own function for #14120 --- modules/scripts.py | 62 ++++++++++++++++++++++++---------------------- 1 file changed, 33 insertions(+), 29 deletions(-) diff --git a/modules/scripts.py b/modules/scripts.py index 961d032ce..7f9454eb5 100644 --- a/modules/scripts.py +++ b/modules/scripts.py @@ -560,54 +560,58 @@ class ScriptRunner: on_after.clear() def create_script_ui(self, script): - import modules.api.models as api_models script.args_from = len(self.inputs) script.args_to = len(self.inputs) + try: + self.create_script_ui_inner(script) + except Exception: + errors.report(f"Error creating UI for {script.name}: ", exc_info=True) + + def create_script_ui_inner(self, script): + import modules.api.models as api_models + controls = wrap_call(script.ui, script.filename, "ui", script.is_img2img) if controls is None: return - try: - script.name = wrap_call(script.title, script.filename, "title", default=script.filename).lower() - api_args = [] + script.name = wrap_call(script.title, script.filename, "title", default=script.filename).lower() - for control in controls: - control.custom_script_source = os.path.basename(script.filename) + api_args = [] - arg_info = api_models.ScriptArg(label=control.label or "") + for control in controls: + control.custom_script_source = os.path.basename(script.filename) - for field in ("value", "minimum", "maximum", "step"): - v = getattr(control, field, None) - if v is not None: - setattr(arg_info, field, v) + arg_info = api_models.ScriptArg(label=control.label or "") - choices = getattr(control, 'choices', None) # as of gradio 3.41, some items in choices are strings, and some are tuples where the first elem is the string - if choices is not None: - arg_info.choices = [x[0] if isinstance(x, tuple) else x for x in choices] + for field in ("value", "minimum", "maximum", "step"): + v = getattr(control, field, None) + if v is not None: + setattr(arg_info, field, v) - api_args.append(arg_info) + choices = getattr(control, 'choices', None) # as of gradio 3.41, some items in choices are strings, and some are tuples where the first elem is the string + if choices is not None: + arg_info.choices = [x[0] if isinstance(x, tuple) else x for x in choices] - script.api_info = api_models.ScriptInfo( - name=script.name, - is_img2img=script.is_img2img, - is_alwayson=script.alwayson, - args=api_args, - ) + api_args.append(arg_info) - if script.infotext_fields is not None: - self.infotext_fields += script.infotext_fields + script.api_info = api_models.ScriptInfo( + name=script.name, + is_img2img=script.is_img2img, + is_alwayson=script.alwayson, + args=api_args, + ) - if script.paste_field_names is not None: - self.paste_field_names += script.paste_field_names + if script.infotext_fields is not None: + self.infotext_fields += script.infotext_fields - self.inputs += controls - script.args_to = len(self.inputs) + if script.paste_field_names is not None: + self.paste_field_names += script.paste_field_names - except Exception: - errors.report(f"Error creating UI for {script.name}: ", exc_info=True) + self.inputs += controls + script.args_to = len(self.inputs) def setup_ui_for_section(self, section, scriptlist=None): if scriptlist is None: From 87cd07b3af74c447b02570bf3963ba83ade2e203 Mon Sep 17 00:00:00 2001 From: Nuullll Date: Sat, 2 Dec 2023 15:54:25 +0800 Subject: [PATCH 112/139] Fix fp64 --- modules/sd_samplers_timesteps_impl.py | 4 ++-- modules/xpu_specific.py | 2 +- 2 files changed, 3 insertions(+), 3 deletions(-) diff --git a/modules/sd_samplers_timesteps_impl.py b/modules/sd_samplers_timesteps_impl.py index a72daafd4..930a64af5 100644 --- a/modules/sd_samplers_timesteps_impl.py +++ b/modules/sd_samplers_timesteps_impl.py @@ -11,7 +11,7 @@ from modules.models.diffusion.uni_pc import uni_pc def ddim(model, x, timesteps, extra_args=None, callback=None, disable=None, eta=0.0): alphas_cumprod = model.inner_model.inner_model.alphas_cumprod alphas = alphas_cumprod[timesteps] - alphas_prev = alphas_cumprod[torch.nn.functional.pad(timesteps[:-1], pad=(1, 0))].to(torch.float64 if x.device.type != 'mps' else torch.float32) + alphas_prev = alphas_cumprod[torch.nn.functional.pad(timesteps[:-1], pad=(1, 0))].to(torch.float64 if x.device.type != 'mps' and x.device.type != 'xpu' else torch.float32) sqrt_one_minus_alphas = torch.sqrt(1 - alphas) sigmas = eta * np.sqrt((1 - alphas_prev.cpu().numpy()) / (1 - alphas.cpu()) * (1 - alphas.cpu() / alphas_prev.cpu().numpy())) @@ -43,7 +43,7 @@ def ddim(model, x, timesteps, extra_args=None, callback=None, disable=None, eta= def plms(model, x, timesteps, extra_args=None, callback=None, disable=None): alphas_cumprod = model.inner_model.inner_model.alphas_cumprod alphas = alphas_cumprod[timesteps] - alphas_prev = alphas_cumprod[torch.nn.functional.pad(timesteps[:-1], pad=(1, 0))].to(torch.float64 if x.device.type != 'mps' else torch.float32) + alphas_prev = alphas_cumprod[torch.nn.functional.pad(timesteps[:-1], pad=(1, 0))].to(torch.float64 if x.device.type != 'mps' and x.device.type != 'xpu' else torch.float32) sqrt_one_minus_alphas = torch.sqrt(1 - alphas) extra_args = {} if extra_args is None else extra_args diff --git a/modules/xpu_specific.py b/modules/xpu_specific.py index 2df68665a..d933c7903 100644 --- a/modules/xpu_specific.py +++ b/modules/xpu_specific.py @@ -4,7 +4,7 @@ from modules.sd_hijack_utils import CondFunc has_ipex = False try: import torch - import intel_extension_for_pytorch as ipex + import intel_extension_for_pytorch as ipex # noqa: F401 has_ipex = True except Exception: pass From 4a666381bf98333ba4512db0f0033df5f6a08771 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Sat, 2 Dec 2023 12:11:21 +0300 Subject: [PATCH 113/139] extras tab batch: actually use original filename preprocessing upscale: do not do an extra upscale step if it's not needed --- modules/postprocessing.py | 4 +++- modules/upscaler.py | 6 +++--- 2 files changed, 6 insertions(+), 4 deletions(-) diff --git a/modules/postprocessing.py b/modules/postprocessing.py index fd0c0cc99..0a134ee43 100644 --- a/modules/postprocessing.py +++ b/modules/postprocessing.py @@ -60,8 +60,10 @@ def run_postprocessing(extras_mode, image, image_folder, input_dir, output_dir, if opts.use_original_name_batch and name is not None: basename = os.path.splitext(os.path.basename(name))[0] + forced_filename = basename else: basename = '' + forced_filename = None infotext = ", ".join([k if k == v else f'{k}: {generation_parameters_copypaste.quote(v)}' for k, v in pp.info.items() if v is not None]) @@ -70,7 +72,7 @@ def run_postprocessing(extras_mode, image, image_folder, input_dir, output_dir, pp.image.info["postprocessing"] = infotext if save_output: - images.save_image(pp.image, path=outpath, basename=basename, seed=None, prompt=None, extension=opts.samples_format, info=infotext, short_filename=True, no_prompt=True, grid=False, pnginfo_section_name="extras", existing_info=existing_pnginfo, forced_filename=None) + images.save_image(pp.image, path=outpath, basename=basename, extension=opts.samples_format, info=infotext, short_filename=True, no_prompt=True, grid=False, pnginfo_section_name="extras", existing_info=existing_pnginfo, forced_filename=forced_filename) if extras_mode != 2 or show_extras_results: outputs.append(pp.image) diff --git a/modules/upscaler.py b/modules/upscaler.py index e682bbaa2..b256e085b 100644 --- a/modules/upscaler.py +++ b/modules/upscaler.py @@ -57,6 +57,9 @@ class Upscaler: dest_h = int((img.height * scale) // 8 * 8) for _ in range(3): + if img.width >= dest_w and img.height >= dest_h: + break + shape = (img.width, img.height) img = self.do_upscale(img, selected_model) @@ -64,9 +67,6 @@ class Upscaler: if shape == (img.width, img.height): break - if img.width >= dest_w and img.height >= dest_h: - break - if img.width != dest_w or img.height != dest_h: img = img.resize((int(dest_w), int(dest_h)), resample=LANCZOS) From 96871e4f744471177d97e01c49f8587d7f67c125 Mon Sep 17 00:00:00 2001 From: Nuullll Date: Sat, 2 Dec 2023 17:11:11 +0800 Subject: [PATCH 114/139] Remove webui-ipex-user.bat --- modules/launch_utils.py | 22 ++++++++++++++++++++++ webui-ipex-user.bat | 19 ------------------- 2 files changed, 22 insertions(+), 19 deletions(-) delete mode 100644 webui-ipex-user.bat diff --git a/modules/launch_utils.py b/modules/launch_utils.py index 264ec9ca6..586cdc7eb 100644 --- a/modules/launch_utils.py +++ b/modules/launch_utils.py @@ -310,6 +310,26 @@ def requirements_met(requirements_file): def prepare_environment(): torch_index_url = os.environ.get('TORCH_INDEX_URL', "https://download.pytorch.org/whl/cu118") torch_command = os.environ.get('TORCH_COMMAND', f"pip install torch==2.0.1 torchvision==0.15.2 --extra-index-url {torch_index_url}") + if args.use_ipex: + if platform.system() == "Windows": + # The "Nuullll/intel-extension-for-pytorch" wheels were built from IPEX source for Intel Arc GPU: https://github.com/intel/intel-extension-for-pytorch/tree/xpu-main + # This is NOT an Intel official release so please use it at your own risk!! + # See https://github.com/Nuullll/intel-extension-for-pytorch/releases/tag/v2.0.110%2Bxpu-master%2Bdll-bundle for details. + # + # Strengths (over official IPEX 2.0.110 windows release): + # - AOT build (for Arc GPU only) to eliminate JIT compilation overhead: https://github.com/intel/intel-extension-for-pytorch/issues/399 + # - Bundles minimal oneAPI 2023.2 dependencies into the python wheels, so users don't need to install oneAPI for the whole system. + # - Provides a compatible torchvision wheel: https://github.com/intel/intel-extension-for-pytorch/issues/465 + # Limitation: + # - Only works for python 3.10 + url_prefix = "https://github.com/Nuullll/intel-extension-for-pytorch/releases/download/v2.0.110%2Bxpu-master%2Bdll-bundle" + torch_command = os.environ.get('TORCH_COMMAND', f"pip install {url_prefix}/torch-2.0.0a0+gite9ebda2-cp310-cp310-win_amd64.whl {url_prefix}/torchvision-0.15.2a0+fa99a53-cp310-cp310-win_amd64.whl {url_prefix}/intel_extension_for_pytorch-2.0.110+gitc6ea20b-cp310-cp310-win_amd64.whl") + else: + # Using official IPEX release for linux since it's already an AOT build. + # However, users still have to install oneAPI toolkit and activate oneAPI environment manually. + # See https://intel.github.io/intel-extension-for-pytorch/index.html#installation for details. + torch_index_url = os.environ.get('TORCH_INDEX_URL', "https://pytorch-extension.intel.com/release-whl/stable/xpu/us/") + torch_command = os.environ.get('TORCH_COMMAND', f"pip install torch==2.0.0a0 intel-extension-for-pytorch==2.0.110+gitba7f6c1 --extra-index-url {torch_index_url}") requirements_file = os.environ.get('REQS_FILE', "requirements_versions.txt") xformers_package = os.environ.get('XFORMERS_PACKAGE', 'xformers==0.0.20') @@ -352,6 +372,8 @@ def prepare_environment(): run(f'"{python}" -m {torch_command}', "Installing torch and torchvision", "Couldn't install torch", live=True) startup_timer.record("install torch") + if args.use_ipex: + args.skip_torch_cuda_test = True if not args.skip_torch_cuda_test and not check_run_python("import torch; assert torch.cuda.is_available()"): raise RuntimeError( 'Torch is not able to use GPU; ' diff --git a/webui-ipex-user.bat b/webui-ipex-user.bat deleted file mode 100644 index ab25a0400..000000000 --- a/webui-ipex-user.bat +++ /dev/null @@ -1,19 +0,0 @@ -@echo off - -set PYTHON= -@REM The "Nuullll/intel-extension-for-pytorch" wheels were built from IPEX source for Intel Arc GPU: https://github.com/intel/intel-extension-for-pytorch/tree/xpu-main -@REM This is NOT an Intel official release so please use it at your own risk!! -@REM See https://github.com/Nuullll/intel-extension-for-pytorch/releases/tag/v2.0.110%2Bxpu-master%2Bdll-bundle for details. -@REM -@REM Strengths (over official IPEX 2.0.110 windows release): -@REM - AOT build (for Arc GPU only) to eliminate JIT compilation overhead: https://github.com/intel/intel-extension-for-pytorch/issues/399 -@REM - Bundles minimal oneAPI 2023.2 dependencies into the python wheels, so users don't need to install oneAPI for the whole system. -@REM - Provides a compatible torchvision wheel: https://github.com/intel/intel-extension-for-pytorch/issues/465 -@REM Limitation: -@REM - Only works for python 3.10 -set "TORCH_COMMAND=pip install https://github.com/Nuullll/intel-extension-for-pytorch/releases/download/v2.0.110%%2Bxpu-master%%2Bdll-bundle/torch-2.0.0a0+gite9ebda2-cp310-cp310-win_amd64.whl https://github.com/Nuullll/intel-extension-for-pytorch/releases/download/v2.0.110%%2Bxpu-master%%2Bdll-bundle/torchvision-0.15.2a0+fa99a53-cp310-cp310-win_amd64.whl https://github.com/Nuullll/intel-extension-for-pytorch/releases/download/v2.0.110%%2Bxpu-master%%2Bdll-bundle/intel_extension_for_pytorch-2.0.110+gitc6ea20b-cp310-cp310-win_amd64.whl" -set GIT= -set VENV_DIR= -set "COMMANDLINE_ARGS=--use-ipex --skip-torch-cuda-test --skip-version-check --opt-sdp-attention" - -call webui.bat From 11d23e8ca55c097ecfa255a05b63f194e25f08be Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Sat, 2 Dec 2023 18:01:11 +0300 Subject: [PATCH 115/139] remove Train/Preprocessing tab and put all its functionality into extras batch images mode --- javascript/ui.js | 17 ++ modules/api/api.py | 15 -- modules/api/models.py | 3 - modules/postprocessing.py | 92 +++++-- modules/scripts_postprocessing.py | 86 ++++++- modules/shared_options.py | 1 + modules/textual_inversion/preprocess.py | 232 ------------------ modules/textual_inversion/ui.py | 7 - modules/ui.py | 107 -------- modules/ui_postprocessing.py | 16 +- modules/ui_toprow.py | 6 +- scripts/postprocessing_caption.py | 30 +++ scripts/postprocessing_codeformer.py | 16 +- .../postprocessing_create_flipped_copies.py | 32 +++ scripts/postprocessing_focal_crop.py | 54 ++++ scripts/postprocessing_gfpgan.py | 13 +- scripts/postprocessing_split_oversized.py | 71 ++++++ scripts/postprocessing_upscale.py | 12 + scripts/processing_autosized_crop.py | 64 +++++ 19 files changed, 460 insertions(+), 414 deletions(-) delete mode 100644 modules/textual_inversion/preprocess.py create mode 100644 scripts/postprocessing_caption.py create mode 100644 scripts/postprocessing_create_flipped_copies.py create mode 100644 scripts/postprocessing_focal_crop.py create mode 100644 scripts/postprocessing_split_oversized.py create mode 100644 scripts/processing_autosized_crop.py diff --git a/javascript/ui.js b/javascript/ui.js index 2e2626020..410fc44e3 100644 --- a/javascript/ui.js +++ b/javascript/ui.js @@ -170,6 +170,23 @@ function submit_img2img() { return res; } +function submit_extras() { + showSubmitButtons('extras', false); + + var id = randomId(); + + requestProgress(id, gradioApp().getElementById('extras_gallery_container'), gradioApp().getElementById('extras_gallery'), function() { + showSubmitButtons('extras', true); + }); + + var res = create_submit_args(arguments); + + res[0] = id; + + console.log(res); + return res; +} + function restoreProgressTxt2img() { showRestoreProgressButton("txt2img", false); var id = localGet("txt2img_task_id"); diff --git a/modules/api/api.py b/modules/api/api.py index 090838747..b3d74e513 100644 --- a/modules/api/api.py +++ b/modules/api/api.py @@ -22,7 +22,6 @@ from modules.api import models from modules.shared import opts from modules.processing import StableDiffusionProcessingTxt2Img, StableDiffusionProcessingImg2Img, process_images from modules.textual_inversion.textual_inversion import create_embedding, train_embedding -from modules.textual_inversion.preprocess import preprocess from modules.hypernetworks.hypernetwork import create_hypernetwork, train_hypernetwork from PIL import PngImagePlugin, Image from modules.sd_models_config import find_checkpoint_config_near_filename @@ -235,7 +234,6 @@ class Api: self.add_api_route("/sdapi/v1/refresh-vae", self.refresh_vae, methods=["POST"]) self.add_api_route("/sdapi/v1/create/embedding", self.create_embedding, methods=["POST"], response_model=models.CreateResponse) self.add_api_route("/sdapi/v1/create/hypernetwork", self.create_hypernetwork, methods=["POST"], response_model=models.CreateResponse) - self.add_api_route("/sdapi/v1/preprocess", self.preprocess, methods=["POST"], response_model=models.PreprocessResponse) self.add_api_route("/sdapi/v1/train/embedding", self.train_embedding, methods=["POST"], response_model=models.TrainResponse) self.add_api_route("/sdapi/v1/train/hypernetwork", self.train_hypernetwork, methods=["POST"], response_model=models.TrainResponse) self.add_api_route("/sdapi/v1/memory", self.get_memory, methods=["GET"], response_model=models.MemoryResponse) @@ -675,19 +673,6 @@ class Api: finally: shared.state.end() - def preprocess(self, args: dict): - try: - shared.state.begin(job="preprocess") - preprocess(**args) # quick operation unless blip/booru interrogation is enabled - shared.state.end() - return models.PreprocessResponse(info='preprocess complete') - except KeyError as e: - return models.PreprocessResponse(info=f"preprocess error: invalid token: {e}") - except Exception as e: - return models.PreprocessResponse(info=f"preprocess error: {e}") - finally: - shared.state.end() - def train_embedding(self, args: dict): try: shared.state.begin(job="train_embedding") diff --git a/modules/api/models.py b/modules/api/models.py index a0d80af8c..33894b3e6 100644 --- a/modules/api/models.py +++ b/modules/api/models.py @@ -202,9 +202,6 @@ class TrainResponse(BaseModel): class CreateResponse(BaseModel): info: str = Field(title="Create info", description="Response string from create embedding or hypernetwork task.") -class PreprocessResponse(BaseModel): - info: str = Field(title="Preprocess info", description="Response string from preprocessing task.") - fields = {} for key, metadata in opts.data_labels.items(): value = opts.data.get(key) diff --git a/modules/postprocessing.py b/modules/postprocessing.py index 0a134ee43..3c85a74c1 100644 --- a/modules/postprocessing.py +++ b/modules/postprocessing.py @@ -6,7 +6,7 @@ from modules import shared, images, devices, scripts, scripts_postprocessing, ui from modules.shared import opts -def run_postprocessing(extras_mode, image, image_folder, input_dir, output_dir, show_extras_results, *args, save_output: bool = True): +def run_postprocessing(id_task, extras_mode, image, image_folder, input_dir, output_dir, show_extras_results, *args, save_output: bool = True): devices.torch_gc() shared.state.begin(job="extras") @@ -29,11 +29,7 @@ def run_postprocessing(extras_mode, image, image_folder, input_dir, output_dir, image_list = shared.listfiles(input_dir) for filename in image_list: - try: - image = Image.open(filename) - except Exception: - continue - yield image, filename + yield filename, filename else: assert image, 'image not selected' yield image, None @@ -45,37 +41,85 @@ def run_postprocessing(extras_mode, image, image_folder, input_dir, output_dir, infotext = '' - for image_data, name in get_images(extras_mode, image, image_folder, input_dir): + data_to_process = list(get_images(extras_mode, image, image_folder, input_dir)) + shared.state.job_count = len(data_to_process) + + for image_placeholder, name in data_to_process: image_data: Image.Image + shared.state.nextjob() shared.state.textinfo = name + shared.state.skipped = False + + if shared.state.interrupted: + break + + if isinstance(image_placeholder, str): + try: + image_data = Image.open(image_placeholder) + except Exception: + continue + else: + image_data = image_placeholder + + shared.state.assign_current_image(image_data) parameters, existing_pnginfo = images.read_info_from_image(image_data) if parameters: existing_pnginfo["parameters"] = parameters - pp = scripts_postprocessing.PostprocessedImage(image_data.convert("RGB")) + initial_pp = scripts_postprocessing.PostprocessedImage(image_data.convert("RGB")) - scripts.scripts_postproc.run(pp, args) + scripts.scripts_postproc.run(initial_pp, args) - if opts.use_original_name_batch and name is not None: - basename = os.path.splitext(os.path.basename(name))[0] - forced_filename = basename - else: - basename = '' - forced_filename = None + if shared.state.skipped: + continue - infotext = ", ".join([k if k == v else f'{k}: {generation_parameters_copypaste.quote(v)}' for k, v in pp.info.items() if v is not None]) + used_suffixes = {} + for pp in [initial_pp, *initial_pp.extra_images]: + suffix = pp.get_suffix(used_suffixes) - if opts.enable_pnginfo: - pp.image.info = existing_pnginfo - pp.image.info["postprocessing"] = infotext + if opts.use_original_name_batch and name is not None: + basename = os.path.splitext(os.path.basename(name))[0] + forced_filename = basename + suffix + else: + basename = '' + forced_filename = None - if save_output: - images.save_image(pp.image, path=outpath, basename=basename, extension=opts.samples_format, info=infotext, short_filename=True, no_prompt=True, grid=False, pnginfo_section_name="extras", existing_info=existing_pnginfo, forced_filename=forced_filename) + infotext = ", ".join([k if k == v else f'{k}: {generation_parameters_copypaste.quote(v)}' for k, v in pp.info.items() if v is not None]) - if extras_mode != 2 or show_extras_results: - outputs.append(pp.image) + if opts.enable_pnginfo: + pp.image.info = existing_pnginfo + pp.image.info["postprocessing"] = infotext + + if save_output: + fullfn, _ = images.save_image(pp.image, path=outpath, basename=basename, extension=opts.samples_format, info=infotext, short_filename=True, no_prompt=True, grid=False, pnginfo_section_name="extras", existing_info=existing_pnginfo, forced_filename=forced_filename, suffix=suffix) + + if pp.caption: + caption_filename = os.path.splitext(fullfn)[0] + ".txt" + if os.path.isfile(caption_filename): + with open(caption_filename, encoding="utf8") as file: + existing_caption = file.read().strip() + else: + existing_caption = "" + + action = shared.opts.postprocessing_existing_caption_action + if action == 'Prepend' and existing_caption: + caption = f"{existing_caption} {pp.caption}" + elif action == 'Append' and existing_caption: + caption = f"{pp.caption} {existing_caption}" + elif action == 'Keep' and existing_caption: + caption = existing_caption + else: + caption = pp.caption + + caption = caption.strip() + if caption: + with open(caption_filename, "w", encoding="utf8") as file: + file.write(caption) + + if extras_mode != 2 or show_extras_results: + outputs.append(pp.image) image_data.close() @@ -99,9 +143,11 @@ def run_extras(extras_mode, resize_mode, image, image_folder, input_dir, output_ "upscaler_2_visibility": extras_upscaler_2_visibility, }, "GFPGAN": { + "enable": True, "gfpgan_visibility": gfpgan_visibility, }, "CodeFormer": { + "enable": True, "codeformer_visibility": codeformer_visibility, "codeformer_weight": codeformer_weight, }, diff --git a/modules/scripts_postprocessing.py b/modules/scripts_postprocessing.py index bac1335dc..901cad080 100644 --- a/modules/scripts_postprocessing.py +++ b/modules/scripts_postprocessing.py @@ -1,13 +1,56 @@ +import dataclasses import os import gradio as gr from modules import errors, shared +@dataclasses.dataclass +class PostprocessedImageSharedInfo: + target_width: int = None + target_height: int = None + + class PostprocessedImage: def __init__(self, image): self.image = image self.info = {} + self.shared = PostprocessedImageSharedInfo() + self.extra_images = [] + self.nametags = [] + self.disable_processing = False + self.caption = None + + def get_suffix(self, used_suffixes=None): + used_suffixes = {} if used_suffixes is None else used_suffixes + suffix = "-".join(self.nametags) + if suffix: + suffix = "-" + suffix + + if suffix not in used_suffixes: + used_suffixes[suffix] = 1 + return suffix + + for i in range(1, 100): + proposed_suffix = suffix + "-" + str(i) + + if proposed_suffix not in used_suffixes: + used_suffixes[proposed_suffix] = 1 + return proposed_suffix + + return suffix + + def create_copy(self, new_image, *, nametags=None, disable_processing=False): + pp = PostprocessedImage(new_image) + pp.shared = self.shared + pp.nametags = self.nametags.copy() + pp.info = self.info.copy() + pp.disable_processing = disable_processing + + if nametags is not None: + pp.nametags += nametags + + return pp class ScriptPostprocessing: @@ -42,10 +85,17 @@ class ScriptPostprocessing: pass - def image_changed(self): + def process_firstpass(self, pp: PostprocessedImage, **args): + """ + Called for all scripts before calling process(). Scripts can examine the image here and set fields + of the pp object to communicate things to other scripts. + args contains a dictionary with all values returned by components from ui() + """ + pass - + def image_changed(self): + pass def wrap_call(func, filename, funcname, *args, default=None, **kwargs): @@ -118,16 +168,42 @@ class ScriptPostprocessingRunner: return inputs def run(self, pp: PostprocessedImage, args): - for script in self.scripts_in_preferred_order(): - shared.state.job = script.name + scripts = [] + for script in self.scripts_in_preferred_order(): script_args = args[script.args_from:script.args_to] process_args = {} for (name, _component), value in zip(script.controls.items(), script_args): process_args[name] = value - script.process(pp, **process_args) + scripts.append((script, process_args)) + + for script, process_args in scripts: + script.process_firstpass(pp, **process_args) + + all_images = [pp] + + for script, process_args in scripts: + if shared.state.skipped: + break + + shared.state.job = script.name + + for single_image in all_images.copy(): + + if not single_image.disable_processing: + script.process(single_image, **process_args) + + for extra_image in single_image.extra_images: + if not isinstance(extra_image, PostprocessedImage): + extra_image = single_image.create_copy(extra_image) + + all_images.append(extra_image) + + single_image.extra_images.clear() + + pp.extra_images = all_images[1:] def create_args_for_run(self, scripts_args): if not self.ui_created: diff --git a/modules/shared_options.py b/modules/shared_options.py index d8a27180e..859dee404 100644 --- a/modules/shared_options.py +++ b/modules/shared_options.py @@ -357,6 +357,7 @@ options_templates.update(options_section(('postprocessing', "Postprocessing", "p 'postprocessing_enable_in_main_ui': OptionInfo([], "Enable postprocessing operations in txt2img and img2img tabs", ui_components.DropdownMulti, lambda: {"choices": [x.name for x in shared_items.postprocessing_scripts()]}), 'postprocessing_operation_order': OptionInfo([], "Postprocessing operation order", ui_components.DropdownMulti, lambda: {"choices": [x.name for x in shared_items.postprocessing_scripts()]}), 'upscaling_max_images_in_cache': OptionInfo(5, "Maximum number of images in upscaling cache", gr.Slider, {"minimum": 0, "maximum": 10, "step": 1}), + 'postprocessing_existing_caption_action': OptionInfo("Ignore", "Action for existing captions", gr.Radio, {"choices": ["Ignore", "Keep", "Prepend", "Append"]}).info("when generating captions using postprocessing; Ignore = use generated; Keep = use original; Prepend/Append = combine both"), })) options_templates.update(options_section((None, "Hidden options"), { diff --git a/modules/textual_inversion/preprocess.py b/modules/textual_inversion/preprocess.py deleted file mode 100644 index 789fa0838..000000000 --- a/modules/textual_inversion/preprocess.py +++ /dev/null @@ -1,232 +0,0 @@ -import os -from PIL import Image, ImageOps -import math -import tqdm - -from modules import shared, images, deepbooru -from modules.textual_inversion import autocrop - - -def preprocess(id_task, process_src, process_dst, process_width, process_height, preprocess_txt_action, process_keep_original_size, process_flip, process_split, process_caption, process_caption_deepbooru=False, split_threshold=0.5, overlap_ratio=0.2, process_focal_crop=False, process_focal_crop_face_weight=0.9, process_focal_crop_entropy_weight=0.15, process_focal_crop_edges_weight=0.5, process_focal_crop_debug=False, process_multicrop=None, process_multicrop_mindim=None, process_multicrop_maxdim=None, process_multicrop_minarea=None, process_multicrop_maxarea=None, process_multicrop_objective=None, process_multicrop_threshold=None): - try: - if process_caption: - shared.interrogator.load() - - if process_caption_deepbooru: - deepbooru.model.start() - - preprocess_work(process_src, process_dst, process_width, process_height, preprocess_txt_action, process_keep_original_size, process_flip, process_split, process_caption, process_caption_deepbooru, split_threshold, overlap_ratio, process_focal_crop, process_focal_crop_face_weight, process_focal_crop_entropy_weight, process_focal_crop_edges_weight, process_focal_crop_debug, process_multicrop, process_multicrop_mindim, process_multicrop_maxdim, process_multicrop_minarea, process_multicrop_maxarea, process_multicrop_objective, process_multicrop_threshold) - - finally: - - if process_caption: - shared.interrogator.send_blip_to_ram() - - if process_caption_deepbooru: - deepbooru.model.stop() - - -def listfiles(dirname): - return os.listdir(dirname) - - -class PreprocessParams: - src = None - dstdir = None - subindex = 0 - flip = False - process_caption = False - process_caption_deepbooru = False - preprocess_txt_action = None - - -def save_pic_with_caption(image, index, params: PreprocessParams, existing_caption=None): - caption = "" - - if params.process_caption: - caption += shared.interrogator.generate_caption(image) - - if params.process_caption_deepbooru: - if caption: - caption += ", " - caption += deepbooru.model.tag_multi(image) - - filename_part = params.src - filename_part = os.path.splitext(filename_part)[0] - filename_part = os.path.basename(filename_part) - - basename = f"{index:05}-{params.subindex}-{filename_part}" - image.save(os.path.join(params.dstdir, f"{basename}.png")) - - if params.preprocess_txt_action == 'prepend' and existing_caption: - caption = f"{existing_caption} {caption}" - elif params.preprocess_txt_action == 'append' and existing_caption: - caption = f"{caption} {existing_caption}" - elif params.preprocess_txt_action == 'copy' and existing_caption: - caption = existing_caption - - caption = caption.strip() - - if caption: - with open(os.path.join(params.dstdir, f"{basename}.txt"), "w", encoding="utf8") as file: - file.write(caption) - - params.subindex += 1 - - -def save_pic(image, index, params, existing_caption=None): - save_pic_with_caption(image, index, params, existing_caption=existing_caption) - - if params.flip: - save_pic_with_caption(ImageOps.mirror(image), index, params, existing_caption=existing_caption) - - -def split_pic(image, inverse_xy, width, height, overlap_ratio): - if inverse_xy: - from_w, from_h = image.height, image.width - to_w, to_h = height, width - else: - from_w, from_h = image.width, image.height - to_w, to_h = width, height - h = from_h * to_w // from_w - if inverse_xy: - image = image.resize((h, to_w)) - else: - image = image.resize((to_w, h)) - - split_count = math.ceil((h - to_h * overlap_ratio) / (to_h * (1.0 - overlap_ratio))) - y_step = (h - to_h) / (split_count - 1) - for i in range(split_count): - y = int(y_step * i) - if inverse_xy: - splitted = image.crop((y, 0, y + to_h, to_w)) - else: - splitted = image.crop((0, y, to_w, y + to_h)) - yield splitted - -# not using torchvision.transforms.CenterCrop because it doesn't allow float regions -def center_crop(image: Image, w: int, h: int): - iw, ih = image.size - if ih / h < iw / w: - sw = w * ih / h - box = (iw - sw) / 2, 0, iw - (iw - sw) / 2, ih - else: - sh = h * iw / w - box = 0, (ih - sh) / 2, iw, ih - (ih - sh) / 2 - return image.resize((w, h), Image.Resampling.LANCZOS, box) - - -def multicrop_pic(image: Image, mindim, maxdim, minarea, maxarea, objective, threshold): - iw, ih = image.size - err = lambda w, h: 1-(lambda x: x if x < 1 else 1/x)(iw/ih/(w/h)) - wh = max(((w, h) for w in range(mindim, maxdim+1, 64) for h in range(mindim, maxdim+1, 64) - if minarea <= w * h <= maxarea and err(w, h) <= threshold), - key= lambda wh: (wh[0]*wh[1], -err(*wh))[::1 if objective=='Maximize area' else -1], - default=None - ) - return wh and center_crop(image, *wh) - - -def preprocess_work(process_src, process_dst, process_width, process_height, preprocess_txt_action, process_keep_original_size, process_flip, process_split, process_caption, process_caption_deepbooru=False, split_threshold=0.5, overlap_ratio=0.2, process_focal_crop=False, process_focal_crop_face_weight=0.9, process_focal_crop_entropy_weight=0.3, process_focal_crop_edges_weight=0.5, process_focal_crop_debug=False, process_multicrop=None, process_multicrop_mindim=None, process_multicrop_maxdim=None, process_multicrop_minarea=None, process_multicrop_maxarea=None, process_multicrop_objective=None, process_multicrop_threshold=None): - width = process_width - height = process_height - src = os.path.abspath(process_src) - dst = os.path.abspath(process_dst) - split_threshold = max(0.0, min(1.0, split_threshold)) - overlap_ratio = max(0.0, min(0.9, overlap_ratio)) - - assert src != dst, 'same directory specified as source and destination' - - os.makedirs(dst, exist_ok=True) - - files = listfiles(src) - - shared.state.job = "preprocess" - shared.state.textinfo = "Preprocessing..." - shared.state.job_count = len(files) - - params = PreprocessParams() - params.dstdir = dst - params.flip = process_flip - params.process_caption = process_caption - params.process_caption_deepbooru = process_caption_deepbooru - params.preprocess_txt_action = preprocess_txt_action - - pbar = tqdm.tqdm(files) - for index, imagefile in enumerate(pbar): - params.subindex = 0 - filename = os.path.join(src, imagefile) - try: - img = Image.open(filename) - img = ImageOps.exif_transpose(img) - img = img.convert("RGB") - except Exception: - continue - - description = f"Preprocessing [Image {index}/{len(files)}]" - pbar.set_description(description) - shared.state.textinfo = description - - params.src = filename - - existing_caption = None - existing_caption_filename = f"{os.path.splitext(filename)[0]}.txt" - if os.path.exists(existing_caption_filename): - with open(existing_caption_filename, 'r', encoding="utf8") as file: - existing_caption = file.read() - - if shared.state.interrupted: - break - - if img.height > img.width: - ratio = (img.width * height) / (img.height * width) - inverse_xy = False - else: - ratio = (img.height * width) / (img.width * height) - inverse_xy = True - - process_default_resize = True - - if process_split and ratio < 1.0 and ratio <= split_threshold: - for splitted in split_pic(img, inverse_xy, width, height, overlap_ratio): - save_pic(splitted, index, params, existing_caption=existing_caption) - process_default_resize = False - - if process_focal_crop and img.height != img.width: - - dnn_model_path = None - try: - dnn_model_path = autocrop.download_and_cache_models() - except Exception as e: - print("Unable to load face detection model for auto crop selection. Falling back to lower quality haar method.", e) - - autocrop_settings = autocrop.Settings( - crop_width = width, - crop_height = height, - face_points_weight = process_focal_crop_face_weight, - entropy_points_weight = process_focal_crop_entropy_weight, - corner_points_weight = process_focal_crop_edges_weight, - annotate_image = process_focal_crop_debug, - dnn_model_path = dnn_model_path, - ) - for focal in autocrop.crop_image(img, autocrop_settings): - save_pic(focal, index, params, existing_caption=existing_caption) - process_default_resize = False - - if process_multicrop: - cropped = multicrop_pic(img, process_multicrop_mindim, process_multicrop_maxdim, process_multicrop_minarea, process_multicrop_maxarea, process_multicrop_objective, process_multicrop_threshold) - if cropped is not None: - save_pic(cropped, index, params, existing_caption=existing_caption) - else: - print(f"skipped {img.width}x{img.height} image {filename} (can't find suitable size within error threshold)") - process_default_resize = False - - if process_keep_original_size: - save_pic(img, index, params, existing_caption=existing_caption) - process_default_resize = False - - if process_default_resize: - img = images.resize_image(1, img, width, height) - save_pic(img, index, params, existing_caption=existing_caption) - - shared.state.nextjob() diff --git a/modules/textual_inversion/ui.py b/modules/textual_inversion/ui.py index 35c4feeff..f149ad1f0 100644 --- a/modules/textual_inversion/ui.py +++ b/modules/textual_inversion/ui.py @@ -3,7 +3,6 @@ import html import gradio as gr import modules.textual_inversion.textual_inversion -import modules.textual_inversion.preprocess from modules import sd_hijack, shared @@ -15,12 +14,6 @@ def create_embedding(name, initialization_text, nvpt, overwrite_old): return gr.Dropdown.update(choices=sorted(sd_hijack.model_hijack.embedding_db.word_embeddings.keys())), f"Created: {filename}", "" -def preprocess(*args): - modules.textual_inversion.preprocess.preprocess(*args) - - return f"Preprocessing {'interrupted' if shared.state.interrupted else 'finished'}.", "" - - def train_embedding(*args): assert not shared.cmd_opts.lowvram, 'Training models with lowvram not possible' diff --git a/modules/ui.py b/modules/ui.py index 08e0ad775..d80486dd4 100644 --- a/modules/ui.py +++ b/modules/ui.py @@ -912,71 +912,6 @@ def create_ui(): with gr.Column(): create_hypernetwork = gr.Button(value="Create hypernetwork", variant='primary', elem_id="train_create_hypernetwork") - with gr.Tab(label="Preprocess images", id="preprocess_images"): - process_src = gr.Textbox(label='Source directory', elem_id="train_process_src") - process_dst = gr.Textbox(label='Destination directory', elem_id="train_process_dst") - process_width = gr.Slider(minimum=64, maximum=2048, step=8, label="Width", value=512, elem_id="train_process_width") - process_height = gr.Slider(minimum=64, maximum=2048, step=8, label="Height", value=512, elem_id="train_process_height") - preprocess_txt_action = gr.Dropdown(label='Existing Caption txt Action', value="ignore", choices=["ignore", "copy", "prepend", "append"], elem_id="train_preprocess_txt_action") - - with gr.Row(): - process_keep_original_size = gr.Checkbox(label='Keep original size', elem_id="train_process_keep_original_size") - process_flip = gr.Checkbox(label='Create flipped copies', elem_id="train_process_flip") - process_split = gr.Checkbox(label='Split oversized images', elem_id="train_process_split") - process_focal_crop = gr.Checkbox(label='Auto focal point crop', elem_id="train_process_focal_crop") - process_multicrop = gr.Checkbox(label='Auto-sized crop', elem_id="train_process_multicrop") - process_caption = gr.Checkbox(label='Use BLIP for caption', elem_id="train_process_caption") - process_caption_deepbooru = gr.Checkbox(label='Use deepbooru for caption', visible=True, elem_id="train_process_caption_deepbooru") - - with gr.Row(visible=False) as process_split_extra_row: - process_split_threshold = gr.Slider(label='Split image threshold', value=0.5, minimum=0.0, maximum=1.0, step=0.05, elem_id="train_process_split_threshold") - process_overlap_ratio = gr.Slider(label='Split image overlap ratio', value=0.2, minimum=0.0, maximum=0.9, step=0.05, elem_id="train_process_overlap_ratio") - - with gr.Row(visible=False) as process_focal_crop_row: - process_focal_crop_face_weight = gr.Slider(label='Focal point face weight', value=0.9, minimum=0.0, maximum=1.0, step=0.05, elem_id="train_process_focal_crop_face_weight") - process_focal_crop_entropy_weight = gr.Slider(label='Focal point entropy weight', value=0.15, minimum=0.0, maximum=1.0, step=0.05, elem_id="train_process_focal_crop_entropy_weight") - process_focal_crop_edges_weight = gr.Slider(label='Focal point edges weight', value=0.5, minimum=0.0, maximum=1.0, step=0.05, elem_id="train_process_focal_crop_edges_weight") - process_focal_crop_debug = gr.Checkbox(label='Create debug image', elem_id="train_process_focal_crop_debug") - - with gr.Column(visible=False) as process_multicrop_col: - gr.Markdown('Each image is center-cropped with an automatically chosen width and height.') - with gr.Row(): - process_multicrop_mindim = gr.Slider(minimum=64, maximum=2048, step=8, label="Dimension lower bound", value=384, elem_id="train_process_multicrop_mindim") - process_multicrop_maxdim = gr.Slider(minimum=64, maximum=2048, step=8, label="Dimension upper bound", value=768, elem_id="train_process_multicrop_maxdim") - with gr.Row(): - process_multicrop_minarea = gr.Slider(minimum=64*64, maximum=2048*2048, step=1, label="Area lower bound", value=64*64, elem_id="train_process_multicrop_minarea") - process_multicrop_maxarea = gr.Slider(minimum=64*64, maximum=2048*2048, step=1, label="Area upper bound", value=640*640, elem_id="train_process_multicrop_maxarea") - with gr.Row(): - process_multicrop_objective = gr.Radio(["Maximize area", "Minimize error"], value="Maximize area", label="Resizing objective", elem_id="train_process_multicrop_objective") - process_multicrop_threshold = gr.Slider(minimum=0, maximum=1, step=0.01, label="Error threshold", value=0.1, elem_id="train_process_multicrop_threshold") - - with gr.Row(): - with gr.Column(scale=3): - gr.HTML(value="") - - with gr.Column(): - with gr.Row(): - interrupt_preprocessing = gr.Button("Interrupt", elem_id="train_interrupt_preprocessing") - run_preprocess = gr.Button(value="Preprocess", variant='primary', elem_id="train_run_preprocess") - - process_split.change( - fn=lambda show: gr_show(show), - inputs=[process_split], - outputs=[process_split_extra_row], - ) - - process_focal_crop.change( - fn=lambda show: gr_show(show), - inputs=[process_focal_crop], - outputs=[process_focal_crop_row], - ) - - process_multicrop.change( - fn=lambda show: gr_show(show), - inputs=[process_multicrop], - outputs=[process_multicrop_col], - ) - def get_textual_inversion_template_names(): return sorted(textual_inversion.textual_inversion_templates) @@ -1077,42 +1012,6 @@ def create_ui(): ] ) - run_preprocess.click( - fn=wrap_gradio_gpu_call(textual_inversion_ui.preprocess, extra_outputs=[gr.update()]), - _js="start_training_textual_inversion", - inputs=[ - dummy_component, - process_src, - process_dst, - process_width, - process_height, - preprocess_txt_action, - process_keep_original_size, - process_flip, - process_split, - process_caption, - process_caption_deepbooru, - process_split_threshold, - process_overlap_ratio, - process_focal_crop, - process_focal_crop_face_weight, - process_focal_crop_entropy_weight, - process_focal_crop_edges_weight, - process_focal_crop_debug, - process_multicrop, - process_multicrop_mindim, - process_multicrop_maxdim, - process_multicrop_minarea, - process_multicrop_maxarea, - process_multicrop_objective, - process_multicrop_threshold, - ], - outputs=[ - ti_output, - ti_outcome, - ], - ) - train_embedding.click( fn=wrap_gradio_gpu_call(textual_inversion_ui.train_embedding, extra_outputs=[gr.update()]), _js="start_training_textual_inversion", @@ -1186,12 +1085,6 @@ def create_ui(): outputs=[], ) - interrupt_preprocessing.click( - fn=lambda: shared.state.interrupt(), - inputs=[], - outputs=[], - ) - loadsave = ui_loadsave.UiLoadsave(cmd_opts.ui_config_file) settings = ui_settings.UiSettings() diff --git a/modules/ui_postprocessing.py b/modules/ui_postprocessing.py index 802e1ce71..fbad0800a 100644 --- a/modules/ui_postprocessing.py +++ b/modules/ui_postprocessing.py @@ -1,9 +1,10 @@ import gradio as gr -from modules import scripts, shared, ui_common, postprocessing, call_queue +from modules import scripts, shared, ui_common, postprocessing, call_queue, ui_toprow import modules.generation_parameters_copypaste as parameters_copypaste def create_ui(): + dummy_component = gr.Label(visible=False) tab_index = gr.State(value=0) with gr.Row(equal_height=False, variant='compact'): @@ -20,11 +21,13 @@ def create_ui(): extras_batch_output_dir = gr.Textbox(label="Output directory", **shared.hide_dirs, placeholder="Leave blank to save images to the default path.", elem_id="extras_batch_output_dir") show_extras_results = gr.Checkbox(label='Show result images', value=True, elem_id="extras_show_extras_results") - submit = gr.Button('Generate', elem_id="extras_generate", variant='primary') - script_inputs = scripts.scripts_postproc.setup_ui() with gr.Column(): + toprow = ui_toprow.Toprow(is_compact=True, is_img2img=False, id_part="extras") + toprow.create_inline_toprow_image() + submit = toprow.submit + result_images, html_info_x, html_info, html_log = ui_common.create_output_panel("extras", shared.opts.outdir_extras_samples) tab_single.select(fn=lambda: 0, inputs=[], outputs=[tab_index]) @@ -33,7 +36,9 @@ def create_ui(): submit.click( fn=call_queue.wrap_gradio_gpu_call(postprocessing.run_postprocessing, extra_outputs=[None, '']), + _js="submit_extras", inputs=[ + dummy_component, tab_index, extras_image, image_batch, @@ -45,8 +50,9 @@ def create_ui(): outputs=[ result_images, html_info_x, - html_info, - ] + html_log, + ], + show_progress=False, ) parameters_copypaste.add_paste_fields("extras", extras_image, None) diff --git a/modules/ui_toprow.py b/modules/ui_toprow.py index 985b5a2dd..88838f977 100644 --- a/modules/ui_toprow.py +++ b/modules/ui_toprow.py @@ -34,8 +34,10 @@ class Toprow: submit_box = None - def __init__(self, is_img2img, is_compact=False): - id_part = "img2img" if is_img2img else "txt2img" + def __init__(self, is_img2img, is_compact=False, id_part=None): + if id_part is None: + id_part = "img2img" if is_img2img else "txt2img" + self.id_part = id_part self.is_img2img = is_img2img self.is_compact = is_compact diff --git a/scripts/postprocessing_caption.py b/scripts/postprocessing_caption.py new file mode 100644 index 000000000..243e3ad9c --- /dev/null +++ b/scripts/postprocessing_caption.py @@ -0,0 +1,30 @@ +from modules import scripts_postprocessing, ui_components, deepbooru, shared +import gradio as gr + + +class ScriptPostprocessingCeption(scripts_postprocessing.ScriptPostprocessing): + name = "Caption" + order = 4000 + + def ui(self): + with ui_components.InputAccordion(False, label="Caption") as enable: + option = gr.CheckboxGroup(value=["Deepbooru"], choices=["Deepbooru", "BLIP"], show_label=False) + + return { + "enable": enable, + "option": option, + } + + def process(self, pp: scripts_postprocessing.PostprocessedImage, enable, option): + if not enable: + return + + captions = [pp.caption] + + if "Deepbooru" in option: + captions.append(deepbooru.model.tag(pp.image)) + + if "BLIP" in option: + captions.append(shared.interrogator.generate_caption(pp.image)) + + pp.caption = ", ".join([x for x in captions if x]) diff --git a/scripts/postprocessing_codeformer.py b/scripts/postprocessing_codeformer.py index a7d80d40e..e1e156ddc 100644 --- a/scripts/postprocessing_codeformer.py +++ b/scripts/postprocessing_codeformer.py @@ -1,28 +1,28 @@ from PIL import Image import numpy as np -from modules import scripts_postprocessing, codeformer_model +from modules import scripts_postprocessing, codeformer_model, ui_components import gradio as gr -from modules.ui_components import FormRow - class ScriptPostprocessingCodeFormer(scripts_postprocessing.ScriptPostprocessing): name = "CodeFormer" order = 3000 def ui(self): - with FormRow(): - codeformer_visibility = gr.Slider(minimum=0.0, maximum=1.0, step=0.001, label="CodeFormer visibility", value=0, elem_id="extras_codeformer_visibility") - codeformer_weight = gr.Slider(minimum=0.0, maximum=1.0, step=0.001, label="CodeFormer weight (0 = maximum effect, 1 = minimum effect)", value=0, elem_id="extras_codeformer_weight") + with ui_components.InputAccordion(False, label="CodeFormer") as enable: + with gr.Row(): + codeformer_visibility = gr.Slider(minimum=0.0, maximum=1.0, step=0.001, label="Visibility", value=1.0, elem_id="extras_codeformer_visibility") + codeformer_weight = gr.Slider(minimum=0.0, maximum=1.0, step=0.001, label="Weight (0 = maximum effect, 1 = minimum effect)", value=0, elem_id="extras_codeformer_weight") return { + "enable": enable, "codeformer_visibility": codeformer_visibility, "codeformer_weight": codeformer_weight, } - def process(self, pp: scripts_postprocessing.PostprocessedImage, codeformer_visibility, codeformer_weight): - if codeformer_visibility == 0: + def process(self, pp: scripts_postprocessing.PostprocessedImage, enable, codeformer_visibility, codeformer_weight): + if codeformer_visibility == 0 or not enable: return restored_img = codeformer_model.codeformer.restore(np.array(pp.image, dtype=np.uint8), w=codeformer_weight) diff --git a/scripts/postprocessing_create_flipped_copies.py b/scripts/postprocessing_create_flipped_copies.py new file mode 100644 index 000000000..3425571dc --- /dev/null +++ b/scripts/postprocessing_create_flipped_copies.py @@ -0,0 +1,32 @@ +from PIL import ImageOps, Image + +from modules import scripts_postprocessing, ui_components +import gradio as gr + + +class ScriptPostprocessingCreateFlippedCopies(scripts_postprocessing.ScriptPostprocessing): + name = "Create flipped copies" + order = 4000 + + def ui(self): + with ui_components.InputAccordion(False, label="Create flipped copies") as enable: + with gr.Row(): + option = gr.CheckboxGroup(value=["Horizontal"], choices=["Horizontal", "Vertical", "Both"], show_label=False) + + return { + "enable": enable, + "option": option, + } + + def process(self, pp: scripts_postprocessing.PostprocessedImage, enable, option): + if not enable: + return + + if "Horizontal" in option: + pp.extra_images.append(ImageOps.mirror(pp.image)) + + if "Vertical" in option: + pp.extra_images.append(pp.image.transpose(Image.Transpose.FLIP_TOP_BOTTOM)) + + if "Both" in option: + pp.extra_images.append(pp.image.transpose(Image.Transpose.FLIP_TOP_BOTTOM).transpose(Image.Transpose.FLIP_LEFT_RIGHT)) diff --git a/scripts/postprocessing_focal_crop.py b/scripts/postprocessing_focal_crop.py new file mode 100644 index 000000000..d3baf2987 --- /dev/null +++ b/scripts/postprocessing_focal_crop.py @@ -0,0 +1,54 @@ + +from modules import scripts_postprocessing, ui_components, errors +import gradio as gr + +from modules.textual_inversion import autocrop + + +class ScriptPostprocessingFocalCrop(scripts_postprocessing.ScriptPostprocessing): + name = "Auto focal point crop" + order = 4000 + + def ui(self): + with ui_components.InputAccordion(False, label="Auto focal point crop") as enable: + face_weight = gr.Slider(label='Focal point face weight', value=0.9, minimum=0.0, maximum=1.0, step=0.05, elem_id="postprocess_focal_crop_face_weight") + entropy_weight = gr.Slider(label='Focal point entropy weight', value=0.15, minimum=0.0, maximum=1.0, step=0.05, elem_id="postprocess_focal_crop_entropy_weight") + edges_weight = gr.Slider(label='Focal point edges weight', value=0.5, minimum=0.0, maximum=1.0, step=0.05, elem_id="postprocess_focal_crop_edges_weight") + debug = gr.Checkbox(label='Create debug image', elem_id="train_process_focal_crop_debug") + + return { + "enable": enable, + "face_weight": face_weight, + "entropy_weight": entropy_weight, + "edges_weight": edges_weight, + "debug": debug, + } + + def process(self, pp: scripts_postprocessing.PostprocessedImage, enable, face_weight, entropy_weight, edges_weight, debug): + if not enable: + return + + if not pp.shared.target_width or not pp.shared.target_height: + return + + dnn_model_path = None + try: + dnn_model_path = autocrop.download_and_cache_models() + except Exception: + errors.report("Unable to load face detection model for auto crop selection. Falling back to lower quality haar method.", exc_info=True) + + autocrop_settings = autocrop.Settings( + crop_width=pp.shared.target_width, + crop_height=pp.shared.target_height, + face_points_weight=face_weight, + entropy_points_weight=entropy_weight, + corner_points_weight=edges_weight, + annotate_image=debug, + dnn_model_path=dnn_model_path, + ) + + result, *others = autocrop.crop_image(pp.image, autocrop_settings) + + pp.image = result + pp.extra_images = [pp.create_copy(x, nametags=["focal-crop-debug"], disable_processing=True) for x in others] + diff --git a/scripts/postprocessing_gfpgan.py b/scripts/postprocessing_gfpgan.py index d854f3f77..6e7566055 100644 --- a/scripts/postprocessing_gfpgan.py +++ b/scripts/postprocessing_gfpgan.py @@ -1,26 +1,25 @@ from PIL import Image import numpy as np -from modules import scripts_postprocessing, gfpgan_model +from modules import scripts_postprocessing, gfpgan_model, ui_components import gradio as gr -from modules.ui_components import FormRow - class ScriptPostprocessingGfpGan(scripts_postprocessing.ScriptPostprocessing): name = "GFPGAN" order = 2000 def ui(self): - with FormRow(): - gfpgan_visibility = gr.Slider(minimum=0.0, maximum=1.0, step=0.001, label="GFPGAN visibility", value=0, elem_id="extras_gfpgan_visibility") + with ui_components.InputAccordion(False, label="GFPGAN") as enable: + gfpgan_visibility = gr.Slider(minimum=0.0, maximum=1.0, step=0.001, label="Visibility", value=1.0, elem_id="extras_gfpgan_visibility") return { + "enable": enable, "gfpgan_visibility": gfpgan_visibility, } - def process(self, pp: scripts_postprocessing.PostprocessedImage, gfpgan_visibility): - if gfpgan_visibility == 0: + def process(self, pp: scripts_postprocessing.PostprocessedImage, enable, gfpgan_visibility): + if gfpgan_visibility == 0 or not enable: return restored_img = gfpgan_model.gfpgan_fix_faces(np.array(pp.image, dtype=np.uint8)) diff --git a/scripts/postprocessing_split_oversized.py b/scripts/postprocessing_split_oversized.py new file mode 100644 index 000000000..c4a03160f --- /dev/null +++ b/scripts/postprocessing_split_oversized.py @@ -0,0 +1,71 @@ +import math + +from modules import scripts_postprocessing, ui_components +import gradio as gr + + +def split_pic(image, inverse_xy, width, height, overlap_ratio): + if inverse_xy: + from_w, from_h = image.height, image.width + to_w, to_h = height, width + else: + from_w, from_h = image.width, image.height + to_w, to_h = width, height + h = from_h * to_w // from_w + if inverse_xy: + image = image.resize((h, to_w)) + else: + image = image.resize((to_w, h)) + + split_count = math.ceil((h - to_h * overlap_ratio) / (to_h * (1.0 - overlap_ratio))) + y_step = (h - to_h) / (split_count - 1) + for i in range(split_count): + y = int(y_step * i) + if inverse_xy: + splitted = image.crop((y, 0, y + to_h, to_w)) + else: + splitted = image.crop((0, y, to_w, y + to_h)) + yield splitted + + +class ScriptPostprocessingSplitOversized(scripts_postprocessing.ScriptPostprocessing): + name = "Split oversized images" + order = 4000 + + def ui(self): + with ui_components.InputAccordion(False, label="Split oversized images") as enable: + with gr.Row(): + split_threshold = gr.Slider(label='Threshold', value=0.5, minimum=0.0, maximum=1.0, step=0.05, elem_id="postprocess_split_threshold") + overlap_ratio = gr.Slider(label='Overlap ratio', value=0.2, minimum=0.0, maximum=0.9, step=0.05, elem_id="postprocess_overlap_ratio") + + return { + "enable": enable, + "split_threshold": split_threshold, + "overlap_ratio": overlap_ratio, + } + + def process(self, pp: scripts_postprocessing.PostprocessedImage, enable, split_threshold, overlap_ratio): + if not enable: + return + + width = pp.shared.target_width + height = pp.shared.target_height + + if not width or not height: + return + + if pp.image.height > pp.image.width: + ratio = (pp.image.width * height) / (pp.image.height * width) + inverse_xy = False + else: + ratio = (pp.image.height * width) / (pp.image.width * height) + inverse_xy = True + + if ratio >= 1.0 and ratio > split_threshold: + return + + result, *others = split_pic(pp.image, inverse_xy, width, height, overlap_ratio) + + pp.image = result + pp.extra_images = [pp.create_copy(x) for x in others] + diff --git a/scripts/postprocessing_upscale.py b/scripts/postprocessing_upscale.py index eb42a29e5..ed709688d 100644 --- a/scripts/postprocessing_upscale.py +++ b/scripts/postprocessing_upscale.py @@ -81,6 +81,14 @@ class ScriptPostprocessingUpscale(scripts_postprocessing.ScriptPostprocessing): return image + def process_firstpass(self, pp: scripts_postprocessing.PostprocessedImage, upscale_mode=1, upscale_by=2.0, upscale_to_width=None, upscale_to_height=None, upscale_crop=False, upscaler_1_name=None, upscaler_2_name=None, upscaler_2_visibility=0.0): + if upscale_mode == 1: + pp.shared.target_width = upscale_to_width + pp.shared.target_height = upscale_to_height + else: + pp.shared.target_width = int(pp.image.width * upscale_by) + pp.shared.target_height = int(pp.image.height * upscale_by) + def process(self, pp: scripts_postprocessing.PostprocessedImage, upscale_mode=1, upscale_by=2.0, upscale_to_width=None, upscale_to_height=None, upscale_crop=False, upscaler_1_name=None, upscaler_2_name=None, upscaler_2_visibility=0.0): if upscaler_1_name == "None": upscaler_1_name = None @@ -126,6 +134,10 @@ class ScriptPostprocessingUpscaleSimple(ScriptPostprocessingUpscale): "upscaler_name": upscaler_name, } + def process_firstpass(self, pp: scripts_postprocessing.PostprocessedImage, upscale_by=2.0, upscaler_name=None): + pp.shared.target_width = int(pp.image.width * upscale_by) + pp.shared.target_height = int(pp.image.height * upscale_by) + def process(self, pp: scripts_postprocessing.PostprocessedImage, upscale_by=2.0, upscaler_name=None): if upscaler_name is None or upscaler_name == "None": return diff --git a/scripts/processing_autosized_crop.py b/scripts/processing_autosized_crop.py new file mode 100644 index 000000000..c09802264 --- /dev/null +++ b/scripts/processing_autosized_crop.py @@ -0,0 +1,64 @@ +from PIL import Image + +from modules import scripts_postprocessing, ui_components +import gradio as gr + + +def center_crop(image: Image, w: int, h: int): + iw, ih = image.size + if ih / h < iw / w: + sw = w * ih / h + box = (iw - sw) / 2, 0, iw - (iw - sw) / 2, ih + else: + sh = h * iw / w + box = 0, (ih - sh) / 2, iw, ih - (ih - sh) / 2 + return image.resize((w, h), Image.Resampling.LANCZOS, box) + + +def multicrop_pic(image: Image, mindim, maxdim, minarea, maxarea, objective, threshold): + iw, ih = image.size + err = lambda w, h: 1 - (lambda x: x if x < 1 else 1 / x)(iw / ih / (w / h)) + wh = max(((w, h) for w in range(mindim, maxdim + 1, 64) for h in range(mindim, maxdim + 1, 64) + if minarea <= w * h <= maxarea and err(w, h) <= threshold), + key=lambda wh: (wh[0] * wh[1], -err(*wh))[::1 if objective == 'Maximize area' else -1], + default=None + ) + return wh and center_crop(image, *wh) + + +class ScriptPostprocessingAutosizedCrop(scripts_postprocessing.ScriptPostprocessing): + name = "Auto-sized crop" + order = 4000 + + def ui(self): + with ui_components.InputAccordion(False, label="Auto-sized crop") as enable: + gr.Markdown('Each image is center-cropped with an automatically chosen width and height.') + with gr.Row(): + mindim = gr.Slider(minimum=64, maximum=2048, step=8, label="Dimension lower bound", value=384, elem_id="postprocess_multicrop_mindim") + maxdim = gr.Slider(minimum=64, maximum=2048, step=8, label="Dimension upper bound", value=768, elem_id="postprocess_multicrop_maxdim") + with gr.Row(): + minarea = gr.Slider(minimum=64 * 64, maximum=2048 * 2048, step=1, label="Area lower bound", value=64 * 64, elem_id="postprocess_multicrop_minarea") + maxarea = gr.Slider(minimum=64 * 64, maximum=2048 * 2048, step=1, label="Area upper bound", value=640 * 640, elem_id="postprocess_multicrop_maxarea") + with gr.Row(): + objective = gr.Radio(["Maximize area", "Minimize error"], value="Maximize area", label="Resizing objective", elem_id="postprocess_multicrop_objective") + threshold = gr.Slider(minimum=0, maximum=1, step=0.01, label="Error threshold", value=0.1, elem_id="postprocess_multicrop_threshold") + + return { + "enable": enable, + "mindim": mindim, + "maxdim": maxdim, + "minarea": minarea, + "maxarea": maxarea, + "objective": objective, + "threshold": threshold, + } + + def process(self, pp: scripts_postprocessing.PostprocessedImage, enable, mindim, maxdim, minarea, maxarea, objective, threshold): + if not enable: + return + + cropped = multicrop_pic(pp.image, mindim, maxdim, minarea, maxarea, objective, threshold) + if cropped is not None: + pp.image = cropped + else: + print(f"skipped {pp.image.width}x{pp.image.height} image (can't find suitable size within error threshold)") From a5f61aa8c5933d8e5a0e0aa841138eeaccd86d62 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Sat, 2 Dec 2023 18:03:34 +0300 Subject: [PATCH 116/139] potential fix for #14172 --- modules/sd_hijack.py | 12 ++++-------- 1 file changed, 4 insertions(+), 8 deletions(-) diff --git a/modules/sd_hijack.py b/modules/sd_hijack.py index 3d340fc9b..14fe62c73 100644 --- a/modules/sd_hijack.py +++ b/modules/sd_hijack.py @@ -38,6 +38,10 @@ ldm.models.diffusion.ddpm.print = shared.ldm_print optimizers = [] current_optimizer: sd_hijack_optimizations.SdOptimization = None +ldm_original_forward = patches.patch(__file__, ldm.modules.diffusionmodules.openaimodel.UNetModel, "forward", sd_unet.UNetModel_forward) +sgm_original_forward = patches.patch(__file__, sgm.modules.diffusionmodules.openaimodel.UNetModel, "forward", sd_unet.UNetModel_forward) + + def list_optimizers(): new_optimizers = script_callbacks.list_optimizers_callback() @@ -255,9 +259,6 @@ class StableDiffusionModelHijack: import modules.models.diffusion.ddpm_edit - ldm_original_forward = patches.patch(__file__, ldm.modules.diffusionmodules.openaimodel.UNetModel, "forward", sd_unet.UNetModel_forward) - sgm_original_forward = patches.patch(__file__, sgm.modules.diffusionmodules.openaimodel.UNetModel, "forward", sd_unet.UNetModel_forward) - if isinstance(m, ldm.models.diffusion.ddpm.LatentDiffusion): sd_unet.original_forward = ldm_original_forward elif isinstance(m, modules.models.diffusion.ddpm_edit.LatentDiffusion): @@ -303,11 +304,6 @@ class StableDiffusionModelHijack: self.layers = None self.clip = None - patches.undo(__file__, ldm.modules.diffusionmodules.openaimodel.UNetModel, "forward") - patches.undo(__file__, sgm.modules.diffusionmodules.openaimodel.UNetModel, "forward") - - sd_unet.original_forward = None - def apply_circular(self, enable): if self.circular_enabled == enable: From ac02216e540cd581f9169c6c791e55721e3117b0 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Sat, 2 Dec 2023 19:35:47 +0300 Subject: [PATCH 117/139] alternate implementation for unet forward replacement that does not depend on hijack being applied --- modules/sd_hijack.py | 7 +++++-- modules/sd_unet.py | 14 ++++++++------ 2 files changed, 13 insertions(+), 8 deletions(-) diff --git a/modules/sd_hijack.py b/modules/sd_hijack.py index 14fe62c73..e139d9964 100644 --- a/modules/sd_hijack.py +++ b/modules/sd_hijack.py @@ -38,8 +38,11 @@ ldm.models.diffusion.ddpm.print = shared.ldm_print optimizers = [] current_optimizer: sd_hijack_optimizations.SdOptimization = None -ldm_original_forward = patches.patch(__file__, ldm.modules.diffusionmodules.openaimodel.UNetModel, "forward", sd_unet.UNetModel_forward) -sgm_original_forward = patches.patch(__file__, sgm.modules.diffusionmodules.openaimodel.UNetModel, "forward", sd_unet.UNetModel_forward) +ldm_patched_forward = sd_unet.create_unet_forward(ldm.modules.diffusionmodules.openaimodel.UNetModel.forward) +ldm_original_forward = patches.patch(__file__, ldm.modules.diffusionmodules.openaimodel.UNetModel, "forward", ldm_patched_forward) + +sgm_patched_forward = sd_unet.create_unet_forward(sgm.modules.diffusionmodules.openaimodel.UNetModel.forward) +sgm_original_forward = patches.patch(__file__, sgm.modules.diffusionmodules.openaimodel.UNetModel, "forward", sgm_patched_forward) def list_optimizers(): diff --git a/modules/sd_unet.py b/modules/sd_unet.py index 6a7bc9e26..a771849c8 100644 --- a/modules/sd_unet.py +++ b/modules/sd_unet.py @@ -5,8 +5,7 @@ from modules import script_callbacks, shared, devices unet_options = [] current_unet_option = None current_unet = None -original_forward = None - +original_forward = None # not used, only left temporarily for compatibility def list_unets(): new_unets = script_callbacks.list_unets_callback() @@ -84,9 +83,12 @@ class SdUnet(torch.nn.Module): pass -def UNetModel_forward(self, x, timesteps=None, context=None, *args, **kwargs): - if current_unet is not None: - return current_unet.forward(x, timesteps, context, *args, **kwargs) +def create_unet_forward(original_forward): + def UNetModel_forward(self, x, timesteps=None, context=None, *args, **kwargs): + if current_unet is not None: + return current_unet.forward(x, timesteps, context, *args, **kwargs) - return original_forward(self, x, timesteps, context, *args, **kwargs) + return original_forward(self, x, timesteps, context, *args, **kwargs) + + return UNetModel_forward From 83e8c322762c545fd589c060811379582926060f Mon Sep 17 00:00:00 2001 From: catboxanon <122327233+catboxanon@users.noreply.github.com> Date: Sat, 2 Dec 2023 13:30:53 -0500 Subject: [PATCH 118/139] Fix `save_samples` being checked early when saving masked composite --- modules/processing.py | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/modules/processing.py b/modules/processing.py index 5ab6dddef..4f265801c 100644 --- a/modules/processing.py +++ b/modules/processing.py @@ -938,14 +938,14 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed: if opts.enable_pnginfo: image.info["parameters"] = text output_images.append(image) - if save_samples and hasattr(p, 'mask_for_overlay') and p.mask_for_overlay and any([opts.save_mask, opts.save_mask_composite, opts.return_mask, opts.return_mask_composite]): + if hasattr(p, 'mask_for_overlay') and p.mask_for_overlay and any([opts.save_mask, opts.save_mask_composite, opts.return_mask, opts.return_mask_composite]): image_mask = p.mask_for_overlay.convert('RGB') image_mask_composite = Image.composite(image.convert('RGBA').convert('RGBa'), Image.new('RGBa', image.size), images.resize_image(2, p.mask_for_overlay, image.width, image.height).convert('L')).convert('RGBA') - if opts.save_mask: + if save_samples and opts.save_mask: images.save_image(image_mask, p.outpath_samples, "", p.seeds[i], p.prompts[i], opts.samples_format, info=infotext(i), p=p, suffix="-mask") - if opts.save_mask_composite: + if save_samples and opts.save_mask_composite: images.save_image(image_mask_composite, p.outpath_samples, "", p.seeds[i], p.prompts[i], opts.samples_format, info=infotext(i), p=p, suffix="-mask-composite") if opts.return_mask: From 9528d66c9479d02c83b8db6107f6b0cb741612dc Mon Sep 17 00:00:00 2001 From: catboxanon <122327233+catboxanon@users.noreply.github.com> Date: Sat, 2 Dec 2023 14:56:26 -0500 Subject: [PATCH 119/139] Re-add setting lost as part of e294e46 --- modules/shared_options.py | 1 + 1 file changed, 1 insertion(+) diff --git a/modules/shared_options.py b/modules/shared_options.py index 859dee404..e5de0d018 100644 --- a/modules/shared_options.py +++ b/modules/shared_options.py @@ -255,6 +255,7 @@ options_templates.update(options_section(('ui_prompt_editing', "Prompt editing", "keyedit_precision_attention": OptionInfo(0.1, "Precision for (attention:1.1) when editing the prompt with Ctrl+up/down", gr.Slider, {"minimum": 0.01, "maximum": 0.2, "step": 0.001}), "keyedit_precision_extra": OptionInfo(0.05, "Precision for when editing the prompt with Ctrl+up/down", gr.Slider, {"minimum": 0.01, "maximum": 0.2, "step": 0.001}), "keyedit_delimiters": OptionInfo(r".,\/!?%^*;:{}=`~() ", "Word delimiters when editing the prompt with Ctrl+up/down"), + "keyedit_delimiters_whitespace": OptionInfo(["Tab", "Carriage Return", "Line Feed"], "Ctrl+up/down whitespace delimiters", gr.CheckboxGroup, lambda: {"choices": ["Tab", "Carriage Return", "Line Feed"]}), "disable_token_counters": OptionInfo(False, "Disable prompt token counters").needs_reload_ui(), })) From d3fdc4af61b7560eede52290e1ede48185680089 Mon Sep 17 00:00:00 2001 From: w-e-w <40751091+w-e-w@users.noreply.github.com> Date: Sun, 3 Dec 2023 18:22:00 +0900 Subject: [PATCH 120/139] rework mask and mask_composite logic --- modules/processing.py | 27 +++++++++++++-------------- 1 file changed, 13 insertions(+), 14 deletions(-) diff --git a/modules/processing.py b/modules/processing.py index 4f265801c..6f01c95f5 100644 --- a/modules/processing.py +++ b/modules/processing.py @@ -938,21 +938,20 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed: if opts.enable_pnginfo: image.info["parameters"] = text output_images.append(image) - if hasattr(p, 'mask_for_overlay') and p.mask_for_overlay and any([opts.save_mask, opts.save_mask_composite, opts.return_mask, opts.return_mask_composite]): - image_mask = p.mask_for_overlay.convert('RGB') - image_mask_composite = Image.composite(image.convert('RGBA').convert('RGBa'), Image.new('RGBa', image.size), images.resize_image(2, p.mask_for_overlay, image.width, image.height).convert('L')).convert('RGBA') + if hasattr(p, 'mask_for_overlay') and p.mask_for_overlay: + if opts.return_mask or opts.save_mask: + image_mask = p.mask_for_overlay.convert('RGB') + if save_samples and opts.save_mask: + images.save_image(image_mask, p.outpath_samples, "", p.seeds[i], p.prompts[i], opts.samples_format, info=infotext(i), p=p, suffix="-mask") + if opts.return_mask: + output_images.append(image_mask) - if save_samples and opts.save_mask: - images.save_image(image_mask, p.outpath_samples, "", p.seeds[i], p.prompts[i], opts.samples_format, info=infotext(i), p=p, suffix="-mask") - - if save_samples and opts.save_mask_composite: - images.save_image(image_mask_composite, p.outpath_samples, "", p.seeds[i], p.prompts[i], opts.samples_format, info=infotext(i), p=p, suffix="-mask-composite") - - if opts.return_mask: - output_images.append(image_mask) - - if opts.return_mask_composite: - output_images.append(image_mask_composite) + if opts.return_mask_composite or opts.save_mask_composite: + image_mask_composite = Image.composite(image.convert('RGBA').convert('RGBa'), Image.new('RGBa', image.size), images.resize_image(2, p.mask_for_overlay, image.width, image.height).convert('L')).convert('RGBA') + if save_samples and opts.save_mask_composite: + images.save_image(image_mask_composite, p.outpath_samples, "", p.seeds[i], p.prompts[i], opts.samples_format, info=infotext(i), p=p, suffix="-mask-composite") + if opts.return_mask_composite: + output_images.append(image_mask_composite) del x_samples_ddim From d92ce145bba714c5b257b9853aa22681233651b8 Mon Sep 17 00:00:00 2001 From: Aarni Koskela Date: Sun, 3 Dec 2023 16:50:20 +0200 Subject: [PATCH 121/139] Add import_hook hack to work around basicsr incompatibility Fixes #13985 --- modules/import_hook.py | 11 +++++++++++ 1 file changed, 11 insertions(+) diff --git a/modules/import_hook.py b/modules/import_hook.py index 28c67dfa8..eba9a3729 100644 --- a/modules/import_hook.py +++ b/modules/import_hook.py @@ -3,3 +3,14 @@ import sys # this will break any attempt to import xformers which will prevent stability diffusion repo from trying to use it if "--xformers" not in "".join(sys.argv): sys.modules["xformers"] = None + +# Hack to fix a changed import in torchvision 0.17+, which otherwise breaks +# basicsr; see https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/13985 +try: + import torchvision.transforms.functional_tensor # noqa: F401 +except ImportError: + try: + import torchvision.transforms.functional as functional + sys.modules["torchvision.transforms.functional_tensor"] = functional + except ImportError: + pass # shrug... From 639ccf254bd4d072f33333abb1ada3d08aaab470 Mon Sep 17 00:00:00 2001 From: illtellyoulater <3078931+illtellyoulater@users.noreply.github.com> Date: Mon, 4 Dec 2023 02:35:35 +0000 Subject: [PATCH 122/139] Update launch_utils.py to fix wrong dep. checks and reinstalls Fixes failing dependency checks for extensions having a different package name and import name (for example ffmpeg-python / ffmpeg), which currently is causing the unneeded reinstall of packages at runtime. In fact with current code, the same string is used when installing a package and when checking for its presence, as you can see in the following example: > launch_utils.run_pip("install ffmpeg-python", "required package") [ Installing required package: "ffmpeg-python" ... ] [ Installed ] > launch_utils.is_installed("ffmpeg-python") False ... which would actually return true with: > launch_utils.is_installed("ffmpeg") True --- modules/launch_utils.py | 14 ++++++++++---- 1 file changed, 10 insertions(+), 4 deletions(-) diff --git a/modules/launch_utils.py b/modules/launch_utils.py index 6e54d0636..6664c5e04 100644 --- a/modules/launch_utils.py +++ b/modules/launch_utils.py @@ -6,6 +6,7 @@ import os import shutil import sys import importlib.util +import importlib.metadata import platform import json from functools import lru_cache @@ -119,11 +120,16 @@ def run(command, desc=None, errdesc=None, custom_env=None, live: bool = default_ def is_installed(package): try: - spec = importlib.util.find_spec(package) - except ModuleNotFoundError: - return False + dist = importlib.metadata.distribution(package) + except importlib.metadata.PackageNotFoundError: + try: + spec = importlib.util.find_spec(package) + except ModuleNotFoundError: + return False - return spec is not None + return spec is not None + + return dist is not None def repo_dir(name): From 06725af40b94a146c56e693a47cbec6d0af55396 Mon Sep 17 00:00:00 2001 From: missionfloyd Date: Sun, 3 Dec 2023 21:26:12 -0700 Subject: [PATCH 123/139] Lint --- modules/launch_utils.py | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/modules/launch_utils.py b/modules/launch_utils.py index 6664c5e04..e71edd01d 100644 --- a/modules/launch_utils.py +++ b/modules/launch_utils.py @@ -120,12 +120,12 @@ def run(command, desc=None, errdesc=None, custom_env=None, live: bool = default_ def is_installed(package): try: - dist = importlib.metadata.distribution(package) + dist = importlib.metadata.distribution(package) except importlib.metadata.PackageNotFoundError: - try: + try: spec = importlib.util.find_spec(package) except ModuleNotFoundError: - return False + return False return spec is not None From 9e1f3feb12a7cfe4fd426dd3df5431c805746ecc Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Mon, 4 Dec 2023 09:15:19 +0300 Subject: [PATCH 124/139] make webui not crash when running with --disable-all-extensions option --- modules/models/diffusion/ddpm_edit.py | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff --git a/modules/models/diffusion/ddpm_edit.py b/modules/models/diffusion/ddpm_edit.py index b892d5fc7..6db340da4 100644 --- a/modules/models/diffusion/ddpm_edit.py +++ b/modules/models/diffusion/ddpm_edit.py @@ -24,10 +24,15 @@ from pytorch_lightning.utilities.distributed import rank_zero_only from ldm.util import log_txt_as_img, exists, default, ismap, isimage, mean_flat, count_params, instantiate_from_config from ldm.modules.ema import LitEma from ldm.modules.distributions.distributions import normal_kl, DiagonalGaussianDistribution -from ldm.models.autoencoder import VQModelInterface, IdentityFirstStage, AutoencoderKL +from ldm.models.autoencoder import IdentityFirstStage, AutoencoderKL from ldm.modules.diffusionmodules.util import make_beta_schedule, extract_into_tensor, noise_like from ldm.models.diffusion.ddim import DDIMSampler +try: + from ldm.models.autoencoder import VQModelInterface +except Exception: + class VQModelInterface: + pass __conditioning_keys__ = {'concat': 'c_concat', 'crossattn': 'c_crossattn', From 48fae7ccdc2fe2d2ba8e8cfcb17b56028734e570 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Mon, 4 Dec 2023 09:35:52 +0300 Subject: [PATCH 125/139] update changelog --- CHANGELOG.md | 162 +++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 162 insertions(+) diff --git a/CHANGELOG.md b/CHANGELOG.md index 2c72359fc..67429bbff 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -1,3 +1,165 @@ +## 1.7.0 + +### Features: +* settings tab rework: add search field, add categories, split UI settings page into many +* add altdiffusion-m18 support ([#13364](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13364)) +* support inference with LyCORIS GLora networks ([#13610](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13610)) +* add lora-embedding bundle system ([#13568](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13568)) +* option to move prompt from top row into generation parameters +* add support for SSD-1B ([#13865](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13865)) +* support inference with OFT networks ([#13692](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13692)) +* script metadata and DAG sorting mechanism ([#13944](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13944)) +* support HyperTile optimization ([#13948](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13948)) +* add support for SD 2.1 Turbo ([#14170](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14170)) +* remove Train->Preprocessing tab and put all its functionality into Extras tab +* initial IPEX support for Intel Arc GPU ([#14171](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14171)) + +### Minor: +* allow reading model hash from images in img2img batch mode ([#12767](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12767)) +* add option to align with sgm repo's sampling implementation ([#12818](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12818)) +* extra field for lora metadata viewer: `ss_output_name` ([#12838](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12838)) +* add action in settings page to calculate all SD checkpoint hashes ([#12909](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12909)) +* add button to copy prompt to style editor ([#12975](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12975)) +* add --skip-load-model-at-start option ([#13253](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13253)) +* write infotext to gif images +* read infotext from gif images ([#13068](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13068)) +* allow configuring the initial state of InputAccordion in ui-config.json ([#13189](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13189)) +* allow editing whitespace delimiters for ctrl+up/ctrl+down prompt editing ([#13444](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13444)) +* prevent accidentally closing popup dialogs ([#13480](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13480)) +* added option to play notification sound or not ([#13631](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13631)) +* show the preview image in the full screen image viewer if available ([#13459](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13459)) +* support for webui.settings.bat ([#13638](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13638)) +* add an option to not print stack traces on ctrl+c +* start/restart generation by Ctrl (Alt) + Enter ([#13644](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13644)) +* update prompts_from_file script to allow concatenating entries with the general prompt ([#13733](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13733)) +* added a visible checkbox to input accordion +* added an option to hide all txt2img/img2img parameters in an accordion ([#13826](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13826)) +* added 'Path' sorting option for Extra network cards ([#13968](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13968)) +* enable prompt hotkeys in style editor ([#13931](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13931)) +* option to show batch img2img results in UI ([#14009](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14009)) +* infotext updates: add option to disregard certain infotext fields, add option to not include VAE in infotext, add explanation to infotext settings page, move some options to infotext settings page +* add FP32 fallback support on sd_vae_approx ([#14046](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14046)) +* support XYZ scripts / split hires path from unet ([#14126](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14126)) +* allow use of mutiple styles csv files ([#14125](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14125)) + +### Extensions and API: +* update gradio to 3.41.2 +* support installed extensions list api ([#12774](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12774)) +* update pnginfo API to return dict with parsed values +* add noisy latent to `ExtraNoiseParams` for callback ([#12856](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12856)) +* show extension datetime in UTC ([#12864](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12864), [#12865](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12865), [#13281](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13281)) +* add an option to choose how to combine hires fix and refiner +* include program version in info response. ([#13135](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13135)) +* sd_unet support for SDXL +* patch DDPM.register_betas so that users can put given_betas in model yaml ([#13276](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13276)) +* xyz_grid: add prepare ([#13266](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13266)) +* allow multiple localization files with same language in extensions ([#13077](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13077)) +* add onEdit function for js and rework token-counter.js to use it +* fix the key error exception when processing override_settings keys ([#13567](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13567)) +* ability for extensions to return custom data via api in response.images ([#13463](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13463)) +* call state.jobnext() before postproces*() ([#13762](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13762)) +* add option to set notification sound volume ([#13884](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13884)) +* update Ruff to 0.1.6 ([#14059](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14059)) +* add Block component creation callback ([#14119](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14119)) +* catch uncaught exception with ui creation scripts ([#14120](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14120)) +* use extension name for determining an extension is installed in the index ([#14063](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14063)) +* update is_installed() from launch_utils.py to fix reinstalling already installed packages ([#14192](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14192)) + +### Bug Fixes: +* fix pix2pix producing bad results +* fix defaults settings page breaking when any of main UI tabs are hidden +* fix error that causes some extra networks to be disabled if both and are present in the prompt +* fix for Reload UI function: if you reload UI on one tab, other opened tabs will no longer stop working +* prevent duplicate resize handler ([#12795](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12795)) +* small typo: vae resolve bug ([#12797](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12797)) +* hide broken image crop tool ([#12792](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12792)) +* don't show hidden samplers in dropdown for XYZ script ([#12780](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12780)) +* fix style editing dialog breaking if it's opened in both img2img and txt2img tabs +* hide --gradio-auth and --api-auth values from /internal/sysinfo report +* add missing infotext for RNG in options ([#12819](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12819)) +* fix notification not playing when built-in webui tab is inactive ([#12834](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12834)) +* honor `--skip-install` for extension installers ([#12832](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12832)) +* don't print blank stdout in extension installers ([#12833](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12833), [#12855](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12855)) +* get progressbar to display correctly in extensions tab +* keep order in list of checkpoints when loading model that doesn't have a checksum +* fix inpainting models in txt2img creating black pictures +* fix generation params regex ([#12876](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12876)) +* fix batch img2img output dir with script ([#12926](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12926)) +* fix #13080 - Hypernetwork/TI preview generation ([#13084](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13084)) +* fix bug with sigma min/max overrides. ([#12995](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12995)) +* more accurate check for enabling cuDNN benchmark on 16XX cards ([#12924](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12924)) +* don't use multicond parser for negative prompt counter ([#13118](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13118)) +* fix data-sort-name containing spaces ([#13412](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13412)) +* update card on correct tab when editing metadata ([#13411](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13411)) +* fix viewing/editing metadata when filename contains an apostrophe ([#13395](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13395)) +* fix: --sd_model in "Prompts from file or textbox" script is not working ([#13302](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13302)) +* better Support for Portable Git ([#13231](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13231)) +* fix issues when webui_dir is not work_dir ([#13210](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13210)) +* fix: lora-bias-backup don't reset cache ([#13178](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13178)) +* account for customizable extra network separators whyen removing extra network text from the prompt ([#12877](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12877)) +* re fix batch img2img output dir with script ([#13170](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13170)) +* fix `--ckpt-dir` path separator and option use `short name` for checkpoint dropdown ([#13139](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13139)) +* consolidated allowed preview formats, Fix extra network `.gif` not woking as preview ([#13121](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13121)) +* fix venv_dir=- environment variable not working as expected on linux ([#13469](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13469)) +* repair unload sd checkpoint button +* edit-attention fixes ([#13533](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13533)) +* fix bug when using --gfpgan-models-path ([#13718](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13718)) +* properly apply sort order for extra network cards when selected from dropdown +* fixes generation restart not working for some users when 'Ctrl+Enter' is pressed ([#13962](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13962)) +* thread safe extra network list_items ([#13014](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13014)) +* fix not able to exit metadata popup when pop up is too big ([#14156](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14156)) +* fix auto focal point crop for opencv >= 4.8 ([#14121](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14121)) +* make 'use-cpu all' actually apply to 'all' ([#14131](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14131)) +* extras tab batch: actually use original filename +* make webui not crash when running with --disable-all-extensions option + +### Other: +* non-local condition ([#12814](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12814)) +* fix minor typos ([#12827](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12827)) +* remove xformers Python version check ([#12842](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12842)) +* style: file-metadata word-break ([#12837](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12837)) +* revert SGM noise multiplier change for img2img because it breaks hires fix +* do not change quicksettings dropdown option when value returned is `None` ([#12854](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12854)) +* [RC 1.6.0 - zoom is partly hidden] Update style.css ([#12839](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12839)) +* chore: change extension time format ([#12851](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12851)) +* WEBUI.SH - Use torch 2.1.0 release candidate for Navi 3 ([#12929](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12929)) +* add Fallback at images.read_info_from_image if exif data was invalid ([#13028](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13028)) +* update cmd arg description ([#12986](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12986)) +* fix: update shared.opts.data when add_option ([#12957](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12957), [#13213](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13213)) +* restore missing tooltips ([#12976](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12976)) +* use default dropdown padding on mobile ([#12880](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12880)) +* put enable console prompts option into settings from commandline args ([#13119](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13119)) +* fix some deprecated types ([#12846](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12846)) +* bump to torchsde==0.2.6 ([#13418](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13418)) +* update dragdrop.js ([#13372](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13372)) +* use orderdict as lru cache:opt/bug ([#13313](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13313)) +* XYZ if not include sub grids do not save sub grid ([#13282](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13282)) +* initialize state.time_start befroe state.job_count ([#13229](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13229)) +* fix fieldname regex ([#13458](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13458)) +* change denoising_strength default to None. ([#13466](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13466)) +* fix regression ([#13475](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13475)) +* fix IndexError ([#13630](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13630)) +* fix: checkpoints_loaded:{checkpoint:state_dict}, model.load_state_dict issue in dict value empty ([#13535](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13535)) +* update bug_report.yml ([#12991](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12991)) +* requirements_versions httpx==0.24.1 ([#13839](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13839)) +* fix parenthesis auto selection ([#13829](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13829)) +* fix #13796 ([#13797](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13797)) +* corrected a typo in `modules/cmd_args.py` ([#13855](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13855)) +* feat: fix randn found element of type float at pos 2 ([#14004](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14004)) +* adds tqdm handler to logging_config.py for progress bar integration ([#13996](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13996)) +* hotfix: call shared.state.end() after postprocessing done ([#13977](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13977)) +* fix dependency address patch 1 ([#13929](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13929)) +* save sysinfo as .json ([#14035](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14035)) +* move exception_records related methods to errors.py ([#14084](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14084)) +* compatibility ([#13936](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13936)) +* json.dump(ensure_ascii=False) ([#14108](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14108)) +* dir buttons start with / so only the correct dir will be shown and no… ([#13957](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/13957)) +* alternate implementation for unet forward replacement that does not depend on hijack being applied +* re-add `keyedit_delimiters_whitespace` setting lost as part of commit e294e46 ([#14178](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14178)) +* fix `save_samples` being checked early when saving masked composite ([#14177](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14177)) +* slight optimization for mask and mask_composite ([#14181](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14181)) +* add import_hook hack to work around basicsr/torchvision incompatibility ([#14186](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14186)) + ## 1.6.1 ### Bug Fixes: From 24dae9bc4cc03a30236957d9c35d37aed79f6f5d Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Mon, 4 Dec 2023 12:36:41 +0300 Subject: [PATCH 126/139] repair old handler for postprocessing API --- modules/postprocessing.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/modules/postprocessing.py b/modules/postprocessing.py index 3c85a74c1..d166f859b 100644 --- a/modules/postprocessing.py +++ b/modules/postprocessing.py @@ -153,4 +153,4 @@ def run_extras(extras_mode, resize_mode, image, image_folder, input_dir, output_ }, }) - return run_postprocessing(extras_mode, image, image_folder, input_dir, output_dir, show_extras_results, *args, save_output=save_output) + return run_postprocessing("", extras_mode, image, image_folder, input_dir, output_dir, show_extras_results, *args, save_output=save_output) From 81105ee0135f1c475920bf44d3a04fc181aed29e Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Mon, 4 Dec 2023 13:11:00 +0300 Subject: [PATCH 127/139] repair old handler for postprocessing API in a way that doesn't break interface --- modules/postprocessing.py | 8 ++++++-- modules/ui_postprocessing.py | 2 +- 2 files changed, 7 insertions(+), 3 deletions(-) diff --git a/modules/postprocessing.py b/modules/postprocessing.py index d166f859b..0c59fad48 100644 --- a/modules/postprocessing.py +++ b/modules/postprocessing.py @@ -6,7 +6,7 @@ from modules import shared, images, devices, scripts, scripts_postprocessing, ui from modules.shared import opts -def run_postprocessing(id_task, extras_mode, image, image_folder, input_dir, output_dir, show_extras_results, *args, save_output: bool = True): +def run_postprocessing(extras_mode, image, image_folder, input_dir, output_dir, show_extras_results, *args, save_output: bool = True): devices.torch_gc() shared.state.begin(job="extras") @@ -128,6 +128,10 @@ def run_postprocessing(id_task, extras_mode, image, image_folder, input_dir, out return outputs, ui_common.plaintext_to_html(infotext), '' +def run_postprocessing_webui(id_task, *args, **kwargs): + return run_postprocessing(*args, **kwargs) + + def run_extras(extras_mode, resize_mode, image, image_folder, input_dir, output_dir, show_extras_results, gfpgan_visibility, codeformer_visibility, codeformer_weight, upscaling_resize, upscaling_resize_w, upscaling_resize_h, upscaling_crop, extras_upscaler_1, extras_upscaler_2, extras_upscaler_2_visibility, upscale_first: bool, save_output: bool = True): """old handler for API""" @@ -153,4 +157,4 @@ def run_extras(extras_mode, resize_mode, image, image_folder, input_dir, output_ }, }) - return run_postprocessing("", extras_mode, image, image_folder, input_dir, output_dir, show_extras_results, *args, save_output=save_output) + return run_postprocessing(extras_mode, image, image_folder, input_dir, output_dir, show_extras_results, *args, save_output=save_output) diff --git a/modules/ui_postprocessing.py b/modules/ui_postprocessing.py index fbad0800a..13d888e48 100644 --- a/modules/ui_postprocessing.py +++ b/modules/ui_postprocessing.py @@ -35,7 +35,7 @@ def create_ui(): tab_batch_dir.select(fn=lambda: 2, inputs=[], outputs=[tab_index]) submit.click( - fn=call_queue.wrap_gradio_gpu_call(postprocessing.run_postprocessing, extra_outputs=[None, '']), + fn=call_queue.wrap_gradio_gpu_call(postprocessing.run_postprocessing_webui, extra_outputs=[None, '']), _js="submit_extras", inputs=[ dummy_component, From 368d66c9ccca0270cc64d6a64d22bfa562f28361 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Mon, 4 Dec 2023 15:56:03 +0300 Subject: [PATCH 128/139] add hypertile infotext --- .../hypertile/scripts/hypertile_script.py | 53 +++++++++++++++---- 1 file changed, 42 insertions(+), 11 deletions(-) diff --git a/extensions-builtin/hypertile/scripts/hypertile_script.py b/extensions-builtin/hypertile/scripts/hypertile_script.py index d3ab60915..395d584b6 100644 --- a/extensions-builtin/hypertile/scripts/hypertile_script.py +++ b/extensions-builtin/hypertile/scripts/hypertile_script.py @@ -17,11 +17,42 @@ class ScriptHypertile(scripts.Script): configure_hypertile(p.width, p.height, enable_unet=shared.opts.hypertile_enable_unet) + self.add_infotext(p) + def before_hr(self, p, *args): + + enable = shared.opts.hypertile_enable_unet_secondpass or shared.opts.hypertile_enable_unet + # exclusive hypertile seed for the second pass - if not shared.opts.hypertile_enable_unet: + if enable: hypertile.set_hypertile_seed(p.all_seeds[0]) - configure_hypertile(p.hr_upscale_to_x, p.hr_upscale_to_y, enable_unet=shared.opts.hypertile_enable_unet_secondpass) + + configure_hypertile(p.hr_upscale_to_x, p.hr_upscale_to_y, enable_unet=enable) + + if enable and not shared.opts.hypertile_enable_unet: + p.extra_generation_params["Hypertile U-Net second pass"] = True + + self.add_infotext(p, add_unet_params=True) + + def add_infotext(self, p, add_unet_params=False): + def option(name): + value = getattr(shared.opts, name) + default_value = shared.opts.get_default(name) + return None if value == default_value else value + + if shared.opts.hypertile_enable_unet: + p.extra_generation_params["Hypertile U-Net"] = True + + if shared.opts.hypertile_enable_unet or add_unet_params: + p.extra_generation_params["Hypertile U-Net max depth"] = option('hypertile_max_depth_unet') + p.extra_generation_params["Hypertile U-Net max tile size"] = option('hypertile_max_tile_unet') + p.extra_generation_params["Hypertile U-Net swap size"] = option('hypertile_swap_size_unet') + + if shared.opts.hypertile_enable_vae: + p.extra_generation_params["Hypertile VAE"] = True + p.extra_generation_params["Hypertile VAE max depth"] = option('hypertile_max_depth_vae') + p.extra_generation_params["Hypertile VAE max tile size"] = option('hypertile_max_tile_vae') + p.extra_generation_params["Hypertile VAE swap size"] = option('hypertile_swap_size_vae') def configure_hypertile(width, height, enable_unet=True): @@ -57,16 +88,16 @@ def on_ui_settings(): benefit. """), - "hypertile_enable_unet": shared.OptionInfo(False, "Enable Hypertile U-Net").info("noticeable change in details of the generated picture; if enabled, overrides the setting below"), - "hypertile_enable_unet_secondpass": shared.OptionInfo(False, "Enable Hypertile U-Net for hires fix second pass"), - "hypertile_max_depth_unet": shared.OptionInfo(3, "Hypertile U-Net max depth", gr.Slider, {"minimum": 0, "maximum": 3, "step": 1}), - "hypertile_max_tile_unet": shared.OptionInfo(256, "Hypertile U-net max tile size", gr.Slider, {"minimum": 0, "maximum": 512, "step": 16}), - "hypertile_swap_size_unet": shared.OptionInfo(3, "Hypertile U-net swap size", gr.Slider, {"minimum": 0, "maximum": 64, "step": 1}), + "hypertile_enable_unet": shared.OptionInfo(False, "Enable Hypertile U-Net", infotext="Hypertile U-Net").info("enables hypertile for all modes, including hires fix second pass; noticeable change in details of the generated picture"), + "hypertile_enable_unet_secondpass": shared.OptionInfo(False, "Enable Hypertile U-Net for hires fix second pass", infotext="Hypertile U-Net second pass").info("enables hypertile just for hires fix second pass - regardless of whether the above setting is enabled"), + "hypertile_max_depth_unet": shared.OptionInfo(3, "Hypertile U-Net max depth", gr.Slider, {"minimum": 0, "maximum": 3, "step": 1}, infotext="Hypertile U-Net max depth").info("larger = more neural network layers affected; minor effect on performance"), + "hypertile_max_tile_unet": shared.OptionInfo(256, "Hypertile U-Net max tile size", gr.Slider, {"minimum": 0, "maximum": 512, "step": 16}, infotext="Hypertile U-Net max tile size").info("larger = worse performance"), + "hypertile_swap_size_unet": shared.OptionInfo(3, "Hypertile U-Net swap size", gr.Slider, {"minimum": 0, "maximum": 64, "step": 1}, infotext="Hypertile U-Net swap size"), - "hypertile_enable_vae": shared.OptionInfo(False, "Enable Hypertile VAE").info("minimal change in the generated picture"), - "hypertile_max_depth_vae": shared.OptionInfo(3, "Hypertile VAE max depth", gr.Slider, {"minimum": 0, "maximum": 3, "step": 1}), - "hypertile_max_tile_vae": shared.OptionInfo(128, "Hypertile VAE max tile size", gr.Slider, {"minimum": 0, "maximum": 512, "step": 16}), - "hypertile_swap_size_vae": shared.OptionInfo(3, "Hypertile VAE swap size ", gr.Slider, {"minimum": 0, "maximum": 64, "step": 1}), + "hypertile_enable_vae": shared.OptionInfo(False, "Enable Hypertile VAE", infotext="Hypertile VAE").info("minimal change in the generated picture"), + "hypertile_max_depth_vae": shared.OptionInfo(3, "Hypertile VAE max depth", gr.Slider, {"minimum": 0, "maximum": 3, "step": 1}, infotext="Hypertile VAE max depth"), + "hypertile_max_tile_vae": shared.OptionInfo(128, "Hypertile VAE max tile size", gr.Slider, {"minimum": 0, "maximum": 512, "step": 16}, infotext="Hypertile VAE max tile size"), + "hypertile_swap_size_vae": shared.OptionInfo(3, "Hypertile VAE swap size ", gr.Slider, {"minimum": 0, "maximum": 64, "step": 1}, infotext="Hypertile VAE swap size"), } for name, opt in options.items(): From 120a84bd2f01ec4489bd12bd68f319798ef30782 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Tue, 5 Dec 2023 07:15:39 +0300 Subject: [PATCH 129/139] Merge pull request #14203 from AUTOMATIC1111/remove-clean_text() remove clean_text() --- modules/styles.py | 23 +++-------------------- 1 file changed, 3 insertions(+), 20 deletions(-) diff --git a/modules/styles.py b/modules/styles.py index 4d218cd7e..7fb6c2e11 100644 --- a/modules/styles.py +++ b/modules/styles.py @@ -2,7 +2,6 @@ import csv import fnmatch import os import os.path -import re import typing import shutil @@ -14,22 +13,6 @@ class PromptStyle(typing.NamedTuple): path: str = None -def clean_text(text: str) -> str: - """ - Iterating through a list of regular expressions and replacement strings, we - clean up the prompt and style text to make it easier to match against each - other. - """ - re_list = [ - ("multiple commas", re.compile("(,+\s+)+,?"), ", "), - ("multiple spaces", re.compile("\s{2,}"), " "), - ] - for _, regex, replace in re_list: - text = regex.sub(replace, text) - - return text.strip(", ") - - def merge_prompts(style_prompt: str, prompt: str) -> str: if "{prompt}" in style_prompt: res = style_prompt.replace("{prompt}", prompt) @@ -44,7 +27,7 @@ def apply_styles_to_prompt(prompt, styles): for style in styles: prompt = merge_prompts(style, prompt) - return clean_text(prompt) + return prompt def unwrap_style_text_from_prompt(style_text, prompt): @@ -56,8 +39,8 @@ def unwrap_style_text_from_prompt(style_text, prompt): Note that the "cleaned" version of the style text is only used for matching purposes here. It isn't returned; the original style text is not modified. """ - stripped_prompt = clean_text(prompt) - stripped_style_text = clean_text(style_text) + stripped_prompt = prompt + stripped_style_text = style_text if "{prompt}" in stripped_style_text: # Work out whether the prompt is wrapped in the style text. If so, we # return True and the "inner" prompt text that isn't part of the style. From 6ef0ff39f2a35a02e5380e522e1dff3eafd7ccfc Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Thu, 14 Dec 2023 09:39:57 +0300 Subject: [PATCH 130/139] Merge pull request #14300 from AUTOMATIC1111/oft_fixes Fix wrong implementation in network_oft --- extensions-builtin/Lora/network_oft.py | 37 ++++++++------------------ 1 file changed, 11 insertions(+), 26 deletions(-) diff --git a/extensions-builtin/Lora/network_oft.py b/extensions-builtin/Lora/network_oft.py index 05c378118..fa647020f 100644 --- a/extensions-builtin/Lora/network_oft.py +++ b/extensions-builtin/Lora/network_oft.py @@ -21,6 +21,8 @@ class NetworkModuleOFT(network.NetworkModule): self.lin_module = None self.org_module: list[torch.Module] = [self.sd_module] + self.scale = 1.0 + # kohya-ss if "oft_blocks" in weights.w.keys(): self.is_kohya = True @@ -53,12 +55,18 @@ class NetworkModuleOFT(network.NetworkModule): self.constraint = None self.block_size, self.num_blocks = factorization(self.out_dim, self.dim) - def calc_updown_kb(self, orig_weight, multiplier): + def calc_updown(self, orig_weight): oft_blocks = self.oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) - oft_blocks = oft_blocks - oft_blocks.transpose(1, 2) # ensure skew-symmetric orthogonal matrix + eye = torch.eye(self.block_size, device=self.oft_blocks.device) + + if self.is_kohya: + block_Q = oft_blocks - oft_blocks.transpose(1, 2) # ensure skew-symmetric orthogonal matrix + norm_Q = torch.norm(block_Q.flatten()) + new_norm_Q = torch.clamp(norm_Q, max=self.constraint) + block_Q = block_Q * ((new_norm_Q + 1e-8) / (norm_Q + 1e-8)) + oft_blocks = torch.matmul(eye + block_Q, (eye - block_Q).float().inverse()) R = oft_blocks.to(orig_weight.device, dtype=orig_weight.dtype) - R = R * multiplier + torch.eye(self.block_size, device=orig_weight.device) # This errors out for MultiheadAttention, might need to be handled up-stream merged_weight = rearrange(orig_weight, '(k n) ... -> k n ...', k=self.num_blocks, n=self.block_size) @@ -72,26 +80,3 @@ class NetworkModuleOFT(network.NetworkModule): updown = merged_weight.to(orig_weight.device, dtype=orig_weight.dtype) - orig_weight output_shape = orig_weight.shape return self.finalize_updown(updown, orig_weight, output_shape) - - def calc_updown(self, orig_weight): - # if alpha is a very small number as in coft, calc_scale() will return a almost zero number so we ignore it - multiplier = self.multiplier() - return self.calc_updown_kb(orig_weight, multiplier) - - # override to remove the multiplier/scale factor; it's already multiplied in get_weight - def finalize_updown(self, updown, orig_weight, output_shape, ex_bias=None): - if self.bias is not None: - updown = updown.reshape(self.bias.shape) - updown += self.bias.to(orig_weight.device, dtype=orig_weight.dtype) - updown = updown.reshape(output_shape) - - if len(output_shape) == 4: - updown = updown.reshape(output_shape) - - if orig_weight.size().numel() == updown.size().numel(): - updown = updown.reshape(orig_weight.shape) - - if ex_bias is not None: - ex_bias = ex_bias * self.multiplier() - - return updown, ex_bias From c7cd9b441d9061f33b7b88be519fb4c6e5b8bc1e Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Thu, 14 Dec 2023 09:41:18 +0300 Subject: [PATCH 131/139] Merge pull request #14296 from akx/paste-resolution Allow pasting in WIDTHxHEIGHT strings into the width/height fields --- javascript/ui.js | 24 ++++++++++++++++++++++++ 1 file changed, 24 insertions(+) diff --git a/javascript/ui.js b/javascript/ui.js index 410fc44e3..18c9f891a 100644 --- a/javascript/ui.js +++ b/javascript/ui.js @@ -215,9 +215,33 @@ function restoreProgressImg2img() { } +/** + * Configure the width and height elements on `tabname` to accept + * pasting of resolutions in the form of "width x height". + */ +function setupResolutionPasting(tabname) { + var width = gradioApp().querySelector(`#${tabname}_width input[type=number]`); + var height = gradioApp().querySelector(`#${tabname}_height input[type=number]`); + for (const el of [width, height]) { + el.addEventListener('paste', function(event) { + var pasteData = event.clipboardData.getData('text/plain'); + var parsed = pasteData.match(/^\s*(\d+)\D+(\d+)\s*$/); + if (parsed) { + width.value = parsed[1]; + height.value = parsed[2]; + updateInput(width); + updateInput(height); + event.preventDefault(); + } + }); + } +} + onUiLoaded(function() { showRestoreProgressButton('txt2img', localGet("txt2img_task_id")); showRestoreProgressButton('img2img', localGet("img2img_task_id")); + setupResolutionPasting('txt2img'); + setupResolutionPasting('img2img'); }); From b55f09c4e13c082590bc64cd792a0b7bd46c1c0d Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Thu, 14 Dec 2023 09:46:05 +0300 Subject: [PATCH 132/139] Merge pull request #14270 from kaalibro/extra-options-elem-id Assign id for "extra_options". Replace numeric field with slider. --- .../extra-options-section/scripts/extra_options_section.py | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/extensions-builtin/extra-options-section/scripts/extra_options_section.py b/extensions-builtin/extra-options-section/scripts/extra_options_section.py index a903df625..ac2c3de46 100644 --- a/extensions-builtin/extra-options-section/scripts/extra_options_section.py +++ b/extensions-builtin/extra-options-section/scripts/extra_options_section.py @@ -23,11 +23,12 @@ class ExtraOptionsSection(scripts.Script): self.setting_names = [] self.infotext_fields = [] extra_options = shared.opts.extra_options_img2img if is_img2img else shared.opts.extra_options_txt2img + elem_id_tabname = "extra_options_" + ("img2img" if is_img2img else "txt2img") mapping = {k: v for v, k in generation_parameters_copypaste.infotext_to_setting_name_mapping} with gr.Blocks() as interface: - with gr.Accordion("Options", open=False) if shared.opts.extra_options_accordion and extra_options else gr.Group(): + with gr.Accordion("Options", open=False, elem_id=elem_id_tabname) if shared.opts.extra_options_accordion and extra_options else gr.Group(elem_id=elem_id_tabname): row_count = math.ceil(len(extra_options) / shared.opts.extra_options_cols) @@ -70,7 +71,7 @@ This page allows you to add some settings to the main interface of txt2img and i """), "extra_options_txt2img": shared.OptionInfo([], "Settings for txt2img", ui_components.DropdownMulti, lambda: {"choices": list(shared.opts.data_labels.keys())}).js("info", "settingsHintsShowQuicksettings").info("setting entries that also appear in txt2img interfaces").needs_reload_ui(), "extra_options_img2img": shared.OptionInfo([], "Settings for img2img", ui_components.DropdownMulti, lambda: {"choices": list(shared.opts.data_labels.keys())}).js("info", "settingsHintsShowQuicksettings").info("setting entries that also appear in img2img interfaces").needs_reload_ui(), - "extra_options_cols": shared.OptionInfo(1, "Number of columns for added settings", gr.Number, {"precision": 0}).needs_reload_ui(), + "extra_options_cols": shared.OptionInfo(1, "Number of columns for added settings", gr.Slider, {"step": 1, "minimum": 1, "maximum": 20}).info("displayed amount will depend on the actual browser window width").needs_reload_ui(), "extra_options_accordion": shared.OptionInfo(False, "Place added settings into an accordion").needs_reload_ui() })) From 888b928f0da7da4a2dfa4519e95ac17a3e5562f7 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Thu, 14 Dec 2023 09:48:14 +0300 Subject: [PATCH 133/139] Merge pull request #14276 from AUTOMATIC1111/fix-styles Fix styles --- modules/styles.py | 31 +++++++------------------------ 1 file changed, 7 insertions(+), 24 deletions(-) diff --git a/modules/styles.py b/modules/styles.py index 7fb6c2e11..81d9800d1 100644 --- a/modules/styles.py +++ b/modules/styles.py @@ -98,10 +98,8 @@ class StyleDatabase: self.path = path folder, file = os.path.split(self.path) - self.default_file = file.split("*")[0] + ".csv" - if self.default_file == ".csv": - self.default_file = "styles.csv" - self.default_path = os.path.join(folder, self.default_file) + filename, _, ext = file.partition('*') + self.default_path = os.path.join(folder, filename + ext) self.prompt_fields = [field for field in PromptStyle._fields if field != "path"] @@ -155,10 +153,8 @@ class StyleDatabase: row["name"], prompt, negative_prompt, path ) - def get_style_paths(self) -> list(): - """ - Returns a list of all distinct paths, including the default path, of - files that styles are loaded from.""" + def get_style_paths(self) -> set: + """Returns a set of all distinct paths of files that styles are loaded from.""" # Update any styles without a path to the default path for style in list(self.styles.values()): if not style.path: @@ -172,9 +168,9 @@ class StyleDatabase: style_paths.add(style.path) # Remove any paths for styles that are just list dividers - style_paths.remove("do_not_save") + style_paths.discard("do_not_save") - return list(style_paths) + return style_paths def get_style_prompts(self, styles): return [self.styles.get(x, self.no_style).prompt for x in styles] @@ -196,20 +192,7 @@ class StyleDatabase: # The path argument is deprecated, but kept for backwards compatibility _ = path - # Update any styles without a path to the default path - for style in list(self.styles.values()): - if not style.path: - self.styles[style.name] = style._replace(path=self.default_path) - - # Create a list of all distinct paths, including the default path - style_paths = set() - style_paths.add(self.default_path) - for _, style in self.styles.items(): - if style.path: - style_paths.add(style.path) - - # Remove any paths for styles that are just list dividers - style_paths.remove("do_not_save") + style_paths = self.get_style_paths() csv_names = [os.path.split(path)[1].lower() for path in style_paths] From 5cb1ce470df8332872af3dfa1067b761062d4608 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Thu, 14 Dec 2023 09:48:36 +0300 Subject: [PATCH 134/139] Merge pull request #14266 from kaalibro/dev Re-add setting lost as part of e294e46 --- modules/shared_options.py | 1 + 1 file changed, 1 insertion(+) diff --git a/modules/shared_options.py b/modules/shared_options.py index e5de0d018..acb6e2d48 100644 --- a/modules/shared_options.py +++ b/modules/shared_options.py @@ -256,6 +256,7 @@ options_templates.update(options_section(('ui_prompt_editing', "Prompt editing", "keyedit_precision_extra": OptionInfo(0.05, "Precision for when editing the prompt with Ctrl+up/down", gr.Slider, {"minimum": 0.01, "maximum": 0.2, "step": 0.001}), "keyedit_delimiters": OptionInfo(r".,\/!?%^*;:{}=`~() ", "Word delimiters when editing the prompt with Ctrl+up/down"), "keyedit_delimiters_whitespace": OptionInfo(["Tab", "Carriage Return", "Line Feed"], "Ctrl+up/down whitespace delimiters", gr.CheckboxGroup, lambda: {"choices": ["Tab", "Carriage Return", "Line Feed"]}), + "keyedit_move": OptionInfo(True, "Alt+left/right moves prompt elements"), "disable_token_counters": OptionInfo(False, "Disable prompt token counters").needs_reload_ui(), })) From b7e0d4a7e171ee1cef73684b8423fe4a20ca7e34 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Thu, 14 Dec 2023 09:52:23 +0300 Subject: [PATCH 135/139] Merge pull request #14229 from Nuullll/ipex-embedding [IPEX] Fix embedding and ControlNet --- modules/xpu_specific.py | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/modules/xpu_specific.py b/modules/xpu_specific.py index d933c7903..d8da94a0e 100644 --- a/modules/xpu_specific.py +++ b/modules/xpu_specific.py @@ -48,3 +48,12 @@ if has_xpu: CondFunc('torch.nn.modules.conv.Conv2d.forward', lambda orig_func, self, input: orig_func(self, input.to(self.weight.data.dtype)), lambda orig_func, self, input: input.dtype != self.weight.data.dtype) + CondFunc('torch.bmm', + lambda orig_func, input, mat2, out=None: orig_func(input.to(mat2.dtype), mat2, out=out), + lambda orig_func, input, mat2, out=None: input.dtype != mat2.dtype) + CondFunc('torch.cat', + lambda orig_func, tensors, dim=0, out=None: orig_func([t.to(tensors[0].dtype) for t in tensors], dim=dim, out=out), + lambda orig_func, tensors, dim=0, out=None: not all(t.dtype == tensors[0].dtype for t in tensors)) + CondFunc('torch.nn.functional.scaled_dot_product_attention', + lambda orig_func, query, key, value, attn_mask=None, dropout_p=0.0, is_causal=False: orig_func(query, key.to(query.dtype), value.to(query.dtype), attn_mask, dropout_p, is_causal), + lambda orig_func, query, key, value, attn_mask=None, dropout_p=0.0, is_causal=False: query.dtype != key.dtype or query.dtype != value.dtype) From f8871dedcfe3a67689ef333aea2fdf05a9aaffa2 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Thu, 14 Dec 2023 09:59:48 +0300 Subject: [PATCH 136/139] Merge pull request #14230 from AUTOMATIC1111/add-option-Live-preview-in-full-page-image-viewer add option: Live preview in full page image viewer --- javascript/imageviewer.js | 2 +- modules/shared_options.py | 1 + 2 files changed, 2 insertions(+), 1 deletion(-) diff --git a/javascript/imageviewer.js b/javascript/imageviewer.js index e4dae91bc..625c5d148 100644 --- a/javascript/imageviewer.js +++ b/javascript/imageviewer.js @@ -34,7 +34,7 @@ function updateOnBackgroundChange() { if (modalImage && modalImage.offsetParent) { let currentButton = selected_gallery_button(); let preview = gradioApp().querySelectorAll('.livePreview > img'); - if (preview.length > 0) { + if (opts.js_live_preview_in_modal_lightbox && preview.length > 0) { // show preview image if available modalImage.src = preview[preview.length - 1].src; } else if (currentButton?.children?.length > 0 && modalImage.src != currentButton.children[0].src) { diff --git a/modules/shared_options.py b/modules/shared_options.py index acb6e2d48..41097d8e6 100644 --- a/modules/shared_options.py +++ b/modules/shared_options.py @@ -331,6 +331,7 @@ options_templates.update(options_section(('ui', "Live previews", "ui"), { "live_preview_content": OptionInfo("Prompt", "Live preview subject", gr.Radio, {"choices": ["Combined", "Prompt", "Negative prompt"]}), "live_preview_refresh_period": OptionInfo(1000, "Progressbar and preview update period").info("in milliseconds"), "live_preview_fast_interrupt": OptionInfo(False, "Return image with chosen live preview method on interrupt").info("makes interrupts faster"), + "js_live_preview_in_modal_lightbox": OptionInfo(True, "Show Live preview in full page image viewer"), })) options_templates.update(options_section(('sampler-params', "Sampler parameters", "sd"), { From eb52c803b849cdd1fc137db4568eca5bb8373f58 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Thu, 14 Dec 2023 10:03:14 +0300 Subject: [PATCH 137/139] Merge pull request #14216 from wfjsw/state-dict-ref-comparison change state dict comparison to ref compare --- modules/sd_disable_initialization.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/modules/sd_disable_initialization.py b/modules/sd_disable_initialization.py index 8863107ae..273a7edd8 100644 --- a/modules/sd_disable_initialization.py +++ b/modules/sd_disable_initialization.py @@ -215,7 +215,7 @@ class LoadStateDictOnMeta(ReplaceHelper): would be on the meta device. """ - if state_dict == sd: + if state_dict is sd: state_dict = {k: v.to(device="meta", dtype=v.dtype) for k, v in state_dict.items()} original(module, state_dict, strict=strict) From 2be85f8fe03533bf3b1ad562ea58ee8227ba3b99 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Thu, 14 Dec 2023 10:08:03 +0300 Subject: [PATCH 138/139] Merge pull request #14237 from ReneKroon/dev #13354 : solve lora loading issue --- extensions-builtin/Lora/networks.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/extensions-builtin/Lora/networks.py b/extensions-builtin/Lora/networks.py index 7f814706a..629bf8537 100644 --- a/extensions-builtin/Lora/networks.py +++ b/extensions-builtin/Lora/networks.py @@ -159,7 +159,8 @@ def load_network(name, network_on_disk): bundle_embeddings = {} for key_network, weight in sd.items(): - key_network_without_network_parts, network_part = key_network.split(".", 1) + key_network_without_network_parts, _, network_part = key_network.partition(".") + if key_network_without_network_parts == "bundle_emb": emb_name, vec_name = network_part.split(".", 1) emb_dict = bundle_embeddings.get(emb_name, {}) From 0dfffe53ec11b2ee097d55efc479f8e707015db9 Mon Sep 17 00:00:00 2001 From: AUTOMATIC1111 <16777216c@gmail.com> Date: Sat, 16 Dec 2023 09:25:08 +0300 Subject: [PATCH 139/139] Merge pull request #14307 from AUTOMATIC1111/default-Falst-js_live_preview_in_modal_lightbox default False js_live_preview_in_modal_lightbox --- modules/shared_options.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/modules/shared_options.py b/modules/shared_options.py index 41097d8e6..d2e86ff10 100644 --- a/modules/shared_options.py +++ b/modules/shared_options.py @@ -331,7 +331,7 @@ options_templates.update(options_section(('ui', "Live previews", "ui"), { "live_preview_content": OptionInfo("Prompt", "Live preview subject", gr.Radio, {"choices": ["Combined", "Prompt", "Negative prompt"]}), "live_preview_refresh_period": OptionInfo(1000, "Progressbar and preview update period").info("in milliseconds"), "live_preview_fast_interrupt": OptionInfo(False, "Return image with chosen live preview method on interrupt").info("makes interrupts faster"), - "js_live_preview_in_modal_lightbox": OptionInfo(True, "Show Live preview in full page image viewer"), + "js_live_preview_in_modal_lightbox": OptionInfo(False, "Show Live preview in full page image viewer"), })) options_templates.update(options_section(('sampler-params', "Sampler parameters", "sd"), {