The content of the image is different from the original

chengzeyi / stable-fast

Best inference performance optimization framework for HuggingFace Diffusers on NVIDIA GPUs.

MIT License

1.05k stars 59 forks source link

The content of the image is different from the original #112

Closed mokcheungyam closed 5 months ago

mokcheungyam commented 5 months ago

Hello! Your project is really impressive! I have tried it on my own machine, but I found that the content of the image is different even when using the same seed, and details are lost. Is there any setting to fix this? :-)

mokcheungyam commented 5 months ago

Oh! Reset "torch.Generator(device="cuda").manual_seed(1)" after wraming up!!! 😂Now I can get the same picture!