MultiDiffusion pipeline now built-in to latest diffusers. Panorama mode enabled. Region-based generation next.

[Alchete]

Some more exciting news in the latent space. MultiDiffusion is now part of the base diffuser library. There are two components to this:

Panorama mode
Region-based text generation (soon)

Panorama mode was just merged and seems like it would only require a checkbox to enable the mode?

The region-based generation will need some drawing and annotating tools, but looks to be very exciting for image composition.

Does someone more in-tune with the tech know if A1111 can simply upgrade its diffusers version to get the panorama merge?

Tweet
https://twitter.com/omerbartal/status/1626442844536819712

Project Page
https://multidiffusion.github.io/

Diffusers merge PR
huggingface/diffusers#2393

Panorama Example

Region-based Example (coming soon)

[catboxanon]

Does someone more in-tune with the tech know if A1111 can simply upgrade its diffusers version to get the panorama merge?

Unfortunately I don't think the pipeline now being part of diffusers is of any worth currently since the webui doesn't utilize diffusers at all. This webui was created right before diffusers started implementing all the SD pipelines. Likely what would need to happen is for a diffusers extension to be created that integrates with the rest of the webui.

Further discussion: #2577 #6585

[Alcyon6]

Vladmandic is adding Diffusers but I don't think Auto is. Since Vladmandic is a fork of Automatic1111 presumably someone could consider pulling the necessary code back over once it's working over there. It seems like most of the cutting edge research stuff comes out for Diffusers first.