stable-diffusion-webui
stable-diffusion-webui copied to clipboard
webui-user error, Waifu Diffusion ckpt
Describe the bug I've tried deleting and reinstalling stable diffusion again and again, want to try if using Waifu diffusion ckpt files would make any difference. It able to pass the Taming Transformers process (a bug when i used the official stable diffusion ckpt, another Issue Report), but now i had a lot more errors, one even containing "The file may be malicious, so the program is not going to read it"
To Reproduce Steps to reproduce the behavior:
- Installing Python 3.10.7 with "Add to PATH" checked
- Installing git with all the recommended options
- Download ckpt file from https://huggingface.co/hakurei/waifu-diffusion-v1-3/tree/main. Saved it in another drive in case I need to recopy it again.
- Download GFPGAN file from https://github.com/TencentARC/GFPGAN. Saved it in another drive in case i need to recopy it again.
- Make a new folder that you want to put the Stable Diffusion on, in the same drive as the Python and git (forgot to document the error that happened when you put it on a different drive. Probably on another bug report), in this case, C:/AI
- Within the folder, right-click and GitBash Here
- Type: git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui
- Making sure that everything is 100% and done. There's one occurrence where there's a process that stops at under 100% and just goes to another process (again, forgot to document), but this aren't it.
- Copy the GFPGAN file to /stable-diffusion-1-4
- Copy the ckpt file to stable-diffusion-1-4/models/Stable Diffusion
- Run the webui-user.bat NOT in Administrator
Expected behavior To smoothly installs all of the ckpts. There haven't even any installing progress bar unlike on many tutorial videos
Screenshots
Desktop (please complete the following information):
- OS: Windows 10 Home 64-bit
- Python : 3.10.7
- AMD Ryzen 7 4800H
- NVIDIA GTX 1660 Ti, 16 GB RAM
Additional context
- There's some page telling to change the ckpt file name into model.ckpt, while also saying that it's only the default and can have multiple ckpt. For this instance of report, I leave the ckpt name as is : model-epoch09-float16.ckpt
- Would the Python 3.10.7 cause this? Since most tutorial out there is using the 3.10.6 version
- I've already set my default graphic cards to NVIDIA, but the Display 1 on my dxdiag is still the one from AMD. Is this the probable cause as well, since i saw on several Issue report that AMD can't
#2235 Seems identical to your problem and it was solved. You may try to do the same steps?
Clear this huggingface dir:
Add "--disable-safe-unpickle" argument to your webui-user.bat file before running it for the first time. It should look like this:
Btw i believe this extra step with putting gfpgan model to root directory is outdated. Every upscale/facefix model will download automatically when you'll use it for the first time (at least it happened for me).
Its certainly not a model issue, nor a model name issue, and probably not a python issue (i use 3.10.7 and it worked with WD and NAI model).
Fixed
Fix solution is as mezotaken wrote :
- Delete C:\user(USER).cache\huggingface
- Edit webui-user.bat, add --disable-safe-unpickle
This successfully creates the local Stable Diffusion. Any ckpt pasted to model folder became an option that can be changed at will

If you didn't choose any ckpt, it will brought up blank black image
This is a workaround not a solution. This problem should not be occurring. I only started getting this exception after git pulling updates today. Something else is wrong here and this should be left open.
I got the same issue. It is found that git lfs is not initialized. So go to https://git-lfs.com/ and install it.
You need to git lfs install once before running the webui.sh