Changing "torch_dtype=torch.float16" to "torch.get_default_dtype()" provides compatibility with the free CPU on Huggingface.co
The renders take hundreds of seconds on the CPU, and generating more than one image at a time causes a connection error.
The image quality is excellent.

Have you tested this?

totes, but only on CPU, so reverse compatibility with GPU would either need more work, or be the original program.
https://huggingface.co/spaces/Omnibus/finetuned_diffusion_cpu

thank you! I just wanted clarification before applying these changes to my duplicate

Ready to merge
This branch is ready to get merged automatically.

Sign up or log in to comment