Schedulers do not work well for text-guided image-to-image generation

[buttercutter]

For https://lablab.ai/t/stable-diffusion-img2img , it seems that only LMSDiscreteScheduler works well in generating different variations of the pictures sketches, that is my observation so far

Besides, it seems that both RePaintScheduler and KarrasVeScheduler are not fully supported yet, at least inside the colab ipynb notebook ? Please advise

Relevant issue at https://github.com/CompVis/stable-diffusion/issues/178

[patrickvonplaten]

Hey @buttercutter,

Note that both repaint and karras_ve are not compatible with stable diffusion.

Would these docs maybe help: https://huggingface.co/docs/diffusers/main/en/using-diffusers/schedulers ?

[buttercutter]

@patrickvonplaten these 2 schedulers are listed in the huggingface link you just sent above. So, why are they not supported in stable diffusion ? I am a bit confused here.

[patrickvonplaten]

@buttercutter neither RePaintScheduler nor KarrasVeScheduler is stated on the doc page: https://huggingface.co/docs/diffusers/main/en/using-diffusers/schedulers -> note that you can always check all compatible schedulers with:

pipeline.scheduler.compatibles

as explained here: https://huggingface.co/docs/diffusers/main/en/using-diffusers/schedulers#changing-the-scheduler

[github-actions[bot]]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.