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

The new smart memory management makes this unnecessary.
comfyanonymous committed 3 years ago
f72780a
Move controlnet code to comfy/controlnet.py
comfyanonymous committed 3 years ago
c77f02e
Move lora code to comfy/lora.py
comfyanonymous committed 3 years ago
15a7716
Move text_projection to base clip model.
comfyanonymous committed 3 years ago
ec96f6d
Code cleanups.
comfyanonymous committed 3 years ago
30eb92c
Try to free enough vram for control lora inference.
comfyanonymous committed 3 years ago
51dde87
e3d0a9a
cc44ade
a6ef08a
Initialize text encoder to target dtype.
comfyanonymous committed 3 years ago
00c0b2c
Don't hardcode node names for image upload widget.
comfyanonymous committed 3 years ago
d7b3b0f
All resolutions now work with t2i adapter for SDXL.
comfyanonymous committed 3 years ago
afcb9cb
T2I adapter SDXL.
comfyanonymous committed 3 years ago
85fde89
Add control lora links to colab notebook.
comfyanonymous committed 3 years ago
f2a7cc9
Controlnet/t2iadapter cleanup.
comfyanonymous committed 3 years ago
cf5ae46
Fix control lora not working in fp32.
comfyanonymous committed 3 years ago
763b0cf
Fix ControlLora on lowvram.
comfyanonymous committed 3 years ago
199d733
Remove autocast from controlnet code.
comfyanonymous committed 3 years ago
d08e53d
remove log
pythongosssss committed 3 years ago
cdaf65c
Small cleanups.
comfyanonymous committed 3 years ago
0d7b0a4
9b1d5a5
Further tuning and fix mem_free_total.
Simon Lui committed 3 years ago
9225465
Free more memory before VAE encode/decode.
comfyanonymous committed 3 years ago
c9b562a