[CamperBob2]

That's what the example code at https://old.reddit.com/r/StableDiffusion/comments/1pqnghp/qw... generates. You get 0.png, 1.png ... n.png, where n= the requested number of layers-1.

It'll drop a 600W RTX 6000 to its knees for about a minute, but it does work.

[dvrp]

I saw some people at a company called Pruna AI got it down to 8 seconds with Cloudflare/Replicate, but I don't know if it was on consumer hardware or an A100/H100/H200, and I don't know if the inference optimization is open-source yet.