Support Stable Diffusion 2

[elixir-nx]

Support Stable Diffusion 2

[seanmor5]

I think these are definitely in scope, it's just a matter of going through those implementations and adding the relevant models and tasks. I am guessing most of the building-blocks are already there with the original stable diffusion implementation, so it should not be very hard!

[zolrath]

I'm not sure if these are beyond the scope of this, but Stable Diffusion supports a number of different modes:

1. text-to-image
2. image-to-image
3. inpainting
4. upscaling
5. depth-to-image.

Currently there is support for text-to-image, but supporting image-to-image and depth-to-image would be great additions without requiring the additional UI something like inpainting would imply. Though, taking a mask image as an additional argument would likely be a good simple path for inpainting.

Depth-to-Image is a volume/shape preserving version of image-to-image introduced in 2.0 which is quite interesting.
https://github.com/Stability-AI/stablediffusion#depth-conditional-stable-diffusion

Would love to see these additional modes supported!