The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.
Commits on master
Don't auto convert clip and vae weights to fp16 when saving checkpoint.
1ddf512 comfyanonymous committed 2 years ago
Save SD3 modelspec.architecture in CheckpointSave node.
32be358 comfyanonymous committed 2 years ago
Load the SD3 T5xxl model in the same dtype stored in the checkpoint.
0e49211 comfyanonymous committed 2 years ago
Update EmptySD3LatentImage to use 1024 resolution by default.
1c34d33 comfyanonymous committed 2 years ago
Require safetensors library to be at least 0.4.2 for fp8 support.
4134564 comfyanonymous committed 2 years ago
Support zeroing out text embeddings with the attention mask.
742d572 comfyanonymous committed 2 years ago
Reshape the empty latent image to the right amount of channels if needed.
6cd8ffc comfyanonymous committed 2 years ago
Add function to get the list of currently loaded models.
104fcea comfyanonymous committed 2 years ago
pytorch xpu should be flash or mem efficient attention?
b1fd26f comfyanonymous committed 2 years ago
Don't load the view coordinates when loading a workflow from the history.
cb8d0eb comfyanonymous committed 2 years ago
Fix to allow use of PerpNegGuider with cfg_function_post hooks (like PAG) (#3618)
e2c585f Peter Crabtree committed 2 years ago