search

ShivamShrirao / diffusers

🤗 Diffusers: State-of-the-art diffusion models for image and audio generation in PyTorch
https://huggingface.co/docs/diffusers
Apache License 2.0
1.89k stars 505 forks source link

version stablize bitsandbytes #178

Closed Luehrsen closed 1 year ago

Luehrsen commented 1 year ago

There has been a degradation in quality for the generated models created with this colab notebook due to an upstream update in the bitsandbytes dependency. This issue has been reported in the Dreambooth Discord and in this repository.

Rolling back to version 0.35.0 of the bitsandbytes dependency appears to resolve the issue and produce results in the expected quality.

See this release yesterday: https://github.com/TimDettmers/bitsandbytes/releases/tag/0.36.0

Thanks to Manders for discovering and testing this issue. All credit to them.

2blackbar commented 1 year ago

I think this or diffusers new code broke entire dreambooth, the results are pretty garbage now , so wheres commit to use 0.35 version ?

ShivamShrirao commented 1 year ago

@2blackbar people had reported it got worse with bitsandbytes update and this fixed it for them.

HarshkumarDegamadiya commented 1 year ago

image

see, my image quality is high as well as 512*512 , but it is giving this output, and when i do prompt it keep showing me this image again and again. i did every change but result are same image