search

FahimF / sd-gui

Clean and simple Stable Diffusion GUI for macOS, Windows, and Linux
MIT License
232 stars 17 forks source link

Integration request: https://github.com/Birch-san/k-diffusion/tree/mps/k_diffusion #11

Open Gitterman69 opened 2 years ago

Gitterman69 commented 2 years ago

Can you integrate this diffusion? It is super optimised!

BIRCH SAN - K DIFFUSION

Gitterman69 commented 2 years ago

Is there a quick way to integrate it myself? can't find the diffusion folders in the folder.... seems the structure is a bit different

FahimF commented 2 years ago

Yeah, Stable Diffusion implementations that use K Diffusion are a totally other beast, sorry 🙂 My implementation is based on Hugging Face Diffusers and so has to use the pipelines and schedulers that Hugging Face provides ... or write my own. This GUI is only about a week or two old and so I am at the point that I'm re-considering the architecture. But probably not till next weekend since I don't get to do a lot of work on this till the weekend.

If I do switch to the other approach then K Diffusion should be simple enough to add in.

Gitterman69 commented 2 years ago

Nice to hear! The only gui alternative on Mac is currently diffusion bee but it is laughable.... no settings at all... BUT it is MPS optimised....

any tips how to get your ui running with other stable diffusion installations on my computer?

FahimF commented 2 years ago

Nothing that would work out of the box I'm afraid. Because the server expects certain parameters and the app will receive responses back from the server. So the parameters have to be named correctly in both cases. And each developer will simply use whatever names they wish. So without taking a look at each one and doing some work it's hard to make it work immediately would be my gut instinct ... without having taken a look at each server in question that is 🙂

I saw your other post about MPS optimizations. There are no optimizations that need to be applied as such. All of the MPS speed ups come from Pytorch. So as long as you have the latest Pytorch nightly (and my code is set to use MPS) then it should all just work.