Note: This model is a Schnell based model, but it requires CFG scale 3.5 or higher (not guidance scale) and 20 steps or more. Needs to be used with clip_l_sumeshi_f1s_v002. (It's the 234.74 MB file in the menu on the right.)
This is an experimental conversion in Spaces using a homebrew script. serverless Inference API does not currently support torch float8_e4m3fn, so it does not work.
I have not been able to confirm if the conversion is working properly.
Please consider this as a test run only.
