* fix AttnProcessor2_0
Fix use of AttnProcessor2_0 for cross attention with mask
* added scale_qk and out_bias flags
* fixed for xformers
* check if it has scale argument
* Update cross_attention.py
* check torch version
* fix sliced attn
* style
* set scale
* fix test
* fixed addedKV processor
* revert back AttnProcessor2_0
* if missing if
* fix inner_dim
---------
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Add support for different model prediction types in DDIMInverseScheduler
Resolve alpha_prod_t_prev index issue for final step of inversion
* Fix old bug introduced when prediction type is "sample"
* Add support for sample clipping for numerical stability and deprecate old kwarg
* Detach sample, alphas, betas
Derive predicted noise from model output before dist. regularization
Style cleanup
* Log loss for debugging
* Revert "Log loss for debugging"
This reverts commit 76ea9c856f99f4c8eca45a0b1801593bb982584b.
* Add comments
* Add inversion equivalence test
* Add expected data for Pix2PixZero pipeline tests with SD 2
* Update tests/pipelines/stable_diffusion/test_stable_diffusion_pix2pix_zero.py
* Remove cruft and add more explanatory comments
---------
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Update README.md
fix 2 bugs: (1) "previous_noisy_sample" should be in the FOR loop in line 87; (2) converting image to INT should be before "Image.fromarray" in line 91
* Apply suggestions from code review
---------
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* fix the in-place modification in unet condition when using controlnet, which will cause backprop errors when training
* add clone to mid block
* fix-copies
---------
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: William Berman <WLBberman@gmail.com>
* ema test cases.
* debugging maessages.
* debugging maessages.
* add: tests for ema.
* fix: optimization_step arg,
* handle device placement.
* Apply suggestions from code review
Co-authored-by: Will Berman <wlbberman@gmail.com>
* remove del and gc.
* address PR feedback.
* add: tests for serialization.
* fix: typos.
* skip_mps to serialization.
---------
Co-authored-by: Will Berman <wlbberman@gmail.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* support for List[ControlNetModel] on init()
* Add to support for multiple ControlNetCondition
* rename conditioning_scale to scale
* scaling bugfix
* Manually merge `MultiControlNet` #2621
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Update src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_controlnet.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Update src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_controlnet.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Update src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_controlnet.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Update src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_controlnet.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Update src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_controlnet.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Update src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_controlnet.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* cleanups
- don't expose ControlNetCondition
- move scaling to ControlNetModel
* make style error correct
* remove ControlNetCondition to reduce code diff
* refactoring image/cond_scale
* add explain for `images`
* Add docstrings
* all fast-test passed
* Add a slow test
* nit
* Apply suggestions from code review
* small precision fix
* nits
MultiControlNet -> MultiControlNetModel - Matches existing naming a bit
closer
MultiControlNetModel inherit from model utils class - Don't have to
re-write fp16 test
Skip tests that save multi controlnet pipeline - Clearer than changing
test body
Don't auto-batch the number of input images to the number of controlnets.
We generally like to require the user to pass the expected number of
inputs. This simplifies the processing code a bit more
Use existing image pre-processing code a bit more. We can rely on the
existing image pre-processing code and keep the inference loop a bit
simpler.
---------
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: William Berman <WLBberman@gmail.com>
* add api doc for flax pipeline + doc string examples
* make style
---------
Co-authored-by: yiyixuxu <yixu@yis-macbook-pro.lan>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Steven Liu