@circfruit It took a while to get this running (4090, 24gb vram, 128gb ram), but I can toss you the ComfyUI workflows I'm using at the moment: https://gist.github.com/Fortyseven/be08c817065b949bca2ba0a2e81dadac

There's some links to the models in the workflow -- that SHOULD do the trick. I also had to make sure I had all of the nodes updated or I'd have various errors even if I had the right models.

I also have these on the Comfy command line: `--use-sage-attention --lowvram --cache-none --reserve-vram 4`. I didn't try tweaking those to see if they were really necessary, so I'm passing them all along just in case. ;)