stable-diffusion-webui
stable-diffusion-webui copied to clipboard
[Bug]: Can no longer generate same exact images upgrading from 1.3.0 to 1.3.1-1.3.2
Is there an existing issue for this?
- [X] I have searched the existing issues and checked the recent builds/commits
What happened?
I have upgraded from AUTOMATIC1111 from 1.3.0 to 1.3.2. when I try to now generate the same images from 1.3.0 with the exact same seed, prompt and everything the images no longer generate the same. I have tried to generate with various images from version 1.3.0 but the result is the same, I can't generate the same image.
Steps to reproduce the problem
- insert 1.3.0 image in png info on automatic1111 v1.3.2
- send to txt2img
- apply same checkpoint/model
- generate image
What should have happened?
Exact same image is generated in the inserted image in png info on Stable diffusion v1.3.2
Commit where the problem happens
https://github.com/AUTOMATIC1111/stable-diffusion-webui/commit/baf6946e06249c5af9851c60171692c44ef633e0
What Python version are you running on ?
Python 3.10.x
What platforms do you use to access the UI ?
Windows
What device are you running WebUI on?
Nvidia GPUs (RTX 20 above)
What browsers do you use to access the UI ?
Mozilla Firefox
Command Line Arguments
No
List of extensions
Stable-Diffusion-Webui-Civitai-Helper sd-dynamic-prompts
Console logs
Normal
Additional information
No response
I have same issue!!
Try the various options in Settings > Compatibility.
Try the various options in Settings > Compatibility.
I have tried the 64 combinations of these and its unable to produce the same image from v1.3.0
Try the various options in Settings > Compatibility.
It didn't work!
This is because the dependencies have been updated. There is nothing developers can do here.
This is because the dependencies have been updated. There is nothing developers can do here.
I don't know what dependencies you are talking about, but the only issue is not just about exact same result. The result of version 3.0 was way better!!
Consistently way better or is it that you're now getting what's effectively a random seed image and comparing it to a cherry-picked image from before?
I also feel the images on 3.0 generated are better than the current version. The changelog should of had a warning that it would be seed breaking at least which it doesn't.
Also if theres no way to fix this how can I downgrade automatic1111 back to 3.0 without a reinstall of sort?
Also if theres no way to fix this how can I downgrade automatic1111 back to 3.0 without a reinstall of sort?
I couldn't downgrade, I mean I've downgraded but the issue still exist!
I haven't checked but if xformers version was updated from 1.3 it will change the output slightly. If 1.3 uses a early version of xformers you could try downgrading to that.