Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Feature Request: Support Differential Diffusion for inpainting. #2671

Closed
throttlekitty opened this issue Jan 29, 2024 · 4 comments · Fixed by #2876
Closed

Feature Request: Support Differential Diffusion for inpainting. #2671

throttlekitty opened this issue Jan 29, 2024 · 4 comments · Fixed by #2876

Comments

@throttlekitty
Copy link

This is a nice alternative to standard inpainting, it allows for the mask to be a gradient for control of strength on top of denoising.

https://github.com/exx8/differential-diffusion

@dfl
Copy link

dfl commented Feb 3, 2024

this looks very cool, but apparently the code hasn’t been released yet.

@throttlekitty
Copy link
Author

True, but it's laid out quite neatly in the paper. I just don't have the chops yet to implement it.

@throttlekitty
Copy link
Author

They've now released their code!
Since the time of my post, I learned of a PR that had been sitting on the automatic1111 repo to perform the same task, I don't know how it compares. AUTOMATIC1111/stable-diffusion-webui#14208

@AlexD81
Copy link

AlexD81 commented Mar 4, 2024

I suppose, it works as denoise strength map , isn't it?
My guess is to use it with "set latent noise mask" but result is not what's expected.
This is with horizontal gradient, but there's no smooth transition, I see 3 discrete levels

изображение
upd. Just realised that smoothess depends on denoising steps. Sadly it won't work well with LCM etc (
Is it possible to fix it somehow to use it with few denoising steps?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging a pull request may close this issue.

3 participants