training-free-structured-diffusion-guidance
training-free-structured-diffusion-guidance copied to clipboard
Using in Stable-Diffusion-gRPCServer
Hi,
Thank you very much for this. I've started integrating it into https://github.com/hafriedlander/stable-diffusion-grpcserver
A couple of notes:
-
I needed to change a couple of calls in StructuredCrossAttention to use _sliced_attention and not just _attention - https://github.com/hafriedlander/stable-diffusion-grpcserver/blob/8582e182c37e32d1cbb353d96cd6f32a7cecb255/sdgrpcserver/pipeline/models/structured_cross_attention.py#L88
-
This code currently doesn't work in FP16 mode without autocast. I will try and fix this soon and then raise a PR.
Thanks again.