Skip to content

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

e7bee85
Fix 7z error when extracting package.
comfyanonymous committed 3 years ago
f5232c4
Fix bug with weights when prompt is long.
comfyanonymous committed 3 years ago
608fcc2
Disable autocast in unet for increased speed.
comfyanonymous committed 3 years ago
ddc6f12
603f02d
Add a conditioning concat node.
comfyanonymous committed 3 years ago
ccb1b25
Support loading unet files in diffusers format.
comfyanonymous committed 3 years ago
af7a499
Properly support SDXL diffusers loras for unet.
comfyanonymous committed 3 years ago
acf9519
Fix size check for conditioning mask
mara committed 3 years ago
c61a95f
Fix issue with OSX.
comfyanonymous committed 3 years ago
8d694cc
c02f3ba
3a09fac
d94ddd8
Pass device to CLIP model.
comfyanonymous committed 3 years ago
c3e96e6
Allow passing custom path to clip-g and clip-h.
comfyanonymous committed 3 years ago
5e6bc82
Improvements for OSX.
comfyanonymous committed 3 years ago
dc9d1f3
Cleanup.
comfyanonymous committed 3 years ago
103c487
Add taesd weights to standalones.
comfyanonymous committed 3 years ago
ae948b4
Switch to fp16 on some cards when the model is too big.
comfyanonymous committed 3 years ago
2c4e0b4
Add a --force-fp16 argument to force fp16 for testing.
comfyanonymous committed 3 years ago
6f3d9f5
--gpu-only now keeps the VAE on the device.
comfyanonymous committed 3 years ago
1c1b0e7
Lower latency by batching some text encoder inputs.
comfyanonymous committed 3 years ago
ce35d8c
Leave text_encoder on the CPU when it can handle it.
comfyanonymous committed 3 years ago
3b6fe51
b6a60fa
Fix nightly packaging.
comfyanonymous committed 3 years ago
fa1959e