James Reynolds

I would love to use ARM64 Brave on my M1 Mac (running Linux).

[TL;DR: The model was trained on mostly 512x512 images, and performs best in regard to coherency at 512x512 generations.](https://www.reddit.com/r/StableDiffusion/comments/wuw9vq/oooook_stablediffusion_should_be_used_in_11_scale/)

I don't have access to the model so I haven't tested it, but based off of what @filipux said, I created [this pull request](https://github.com/CompVis/stable-diffusion/pull/47) to add mps support. If you...

@Raymonf: I merged your changes with mine and so they are in the pull request now. It caught everything that I missed and it almost identical to the changes that...

> mps support for aten::index.Tensor_out is now in pytorch nightly according to Denis Is that the pytorch nightly branch? That particular branch is 1068 commits ahead and 28606 commits behind...

I changed it back to the old way but then I check if the file exists, if debug is a table, and if debug.getinfo is a function before using it.

I should add that ljcacioppo gets around this by making [most of the values optional](https://github.com/Jamf-Custom-Profile-Schemas/ljcacioppo-schemas/blob/main/menu.nomad.login.ad.json). Maybe something similar could be done here in the build script?

Please use [this fork/branch](https://github.com/lstein/stable-diffusion/tree/magnusviri-apple-mps-support).

It looks like [this is the solution](https://support.apple.com/en-us/HT207112).

@JustinGuese, would you be willing to add this to https://github.com/lstein/stable-diffusion?