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.
f72780a comfyanonymous committed 3 years ago
Fix potential issue with text projection matrix multiplication.
e3d0a9a comfyanonymous committed 3 years ago
Always shift text encoder to GPU when the device supports fp16.
cc44ade comfyanonymous committed 3 years ago
Even with forced fp16 the cpu device should never use it.
a6ef08a comfyanonymous committed 3 years ago
Save memory by storing text encoder weights in fp16 in most situations.
f081017 comfyanonymous committed 3 years ago
Add clip_vision_g download command to colab notebook for ReVision.
e2256b4 comfyanonymous committed 3 years ago
Merge branch 'custom-node-js' of https://github.com/pythongosssss/ComfyUI
bc76b38 comfyanonymous committed 3 years ago
Allow loading js extensions without copying to /web folder
9b1d5a5 pythongosssss committed 3 years ago
Add ipex optimize and other enhancements for Intel GPUs based on recent memory changes.
2c096e4 Simon Lui committed 3 years ago
Merge branch 'parallel-extensions-load' of https://github.com/NoCrypt/ComfyUI
8ee0473 comfyanonymous committed 3 years ago
--disable-smart-memory now disables loading model directly to vram.
e9469e7 comfyanonymous committed 3 years ago
Make the extensions loads in parallel instead of waiting one by one
81ccaca ncpt committed 3 years ago