comfyanonymous
comfyanonymous
The original stable diffusion implementation for SD1.x only supports latent sizes that are multiples of 8 however there is a trick ([here it is implemented in diffusers](https://github.com/huggingface/diffusers/blob/160474ac61934cc22793d6cebea118c171175dbc/src/diffusers/models/resnet.py#L160)) to upsample the...
# 🐛 Bug Xformers gives a CUDA error like this when the batch size is larger or equal to 65536. ``` RuntimeError: CUDA error: invalid configuration argument CUDA kernel errors...
This fixes a copy paste bug where some links are sometimes pasted incorrectly when copy pasting partial graphs.
I'm leaving this here because I'm not sure if I implemented it correctly or if the results are supposed to be this bad. It only seems to work with FluxGuidance...
These changes makes the memory management less fragile (much less chances of custom nodes/extensions/future code changes breaking it) and should remove the noticeable delay when changing workflows with large models....