Mirrors
/
EveryDream-trainer
mirror of https://github.com/victorchall/EveryDream-trainer.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Update README.md

This commit is contained in:
JoePenna 2022-09-29 07:28:59 -07:00 committed by GitHub
parent 3032396587
commit c99eb2b8e0
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 20 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
README.md
View File

@ -4,6 +4,7 @@
- [Setup](#setup) - [Setup](#setup)
- [Easy RunPod Instructions](#easy-runpod-instructions) - [Easy RunPod Instructions](#easy-runpod-instructions)
- [Vast.AI Setup](#vast-ai-setup) - [Vast.AI Setup](#vast-ai-setup)
- [Textual Inversion vs. Dreambooth](#text-vs-dreamb)
- [Using the Generated Model](#using-the-generated-model) - [Using the Generated Model](#using-the-generated-model)
- [Debugging Your Results](#debugging-your-results) - [Debugging Your Results](#debugging-your-results)
- [They don't look like you at all!](#they-dont-look-like-you) - [They don't look like you at all!](#they-dont-look-like-you)
@ -12,7 +13,7 @@
- [Hugging Face Diffusers](#hugging-face-diffusers) - [Hugging Face Diffusers](#hugging-face-diffusers)
# The Repo Formerly Known As "Dreambooth" # The Repo Formerly Known As "Dreambooth"
## ...but is more correctly described as "Unfrozen Model Textual Inversion for Stable Diffusion" ## ...now more accurately described as "Unfrozen Model Textual Inversion for Stable Diffusion"
![image](https://user-images.githubusercontent.com/100188076/192390551-cb89364f-af57-4aed-8f3d-f9eb9b61cf95.png) ![image](https://user-images.githubusercontent.com/100188076/192390551-cb89364f-af57-4aed-8f3d-f9eb9b61cf95.png)
## <a name="notes-by-joe-penna"></a> Notes by Joe Penna ## <a name="notes-by-joe-penna"></a> Notes by Joe Penna
@ -85,6 +86,24 @@ Now, if you wanna try to do this... please read the warnings below first:
- ![img.png](readme-images/vast-ai-step6-open-notebook.png) - ![img.png](readme-images/vast-ai-step6-open-notebook.png)
- Follow the instructions in the workbook and start training - Follow the instructions in the workbook and start training
# <a name="text-vs-dreamb"></a> Textual Inversion vs. Dreambooth
The majority of the code in this repo was written by Rinon Gal et. al, the authors of the Textual Inversion research paper.
A few bits about regularization images were added that we all thought were super important -- all the researchers included!
...until my images were trained under the class "dog":
<br><img src="https://media.discordapp.net/attachments/1024716296610385981/1024933960083587102/unknown.png" width="200">
...and under the nonsensical class "§¶•" instead of "man" or "woman" or "person":
<br><img src="https://media.discordapp.net/attachments/1024716296610385981/1024934146415529984/unknown.png" width="200">
...and with completely blank regularization images:
<br><img src="https://media.discordapp.net/attachments/1023293330601287711/1024933371102629898/IMG_7579.JPG" width="200">
So, because regularization has no effect, this isn't Dreambooth at all.
This repo is more accurately called *"Unfrozen Model Textual Inversion for Stable Diffusion"*.
# <a name="using-the-generated-model"></a> Using the generated model # <a name="using-the-generated-model"></a> Using the generated model
The `ground truth` (real picture, caution: very beautiful woman) The `ground truth` (real picture, caution: very beautiful woman)