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
Fix LTXVPreprocess on resolutions that are not multiples of 2.
30e6cfb comfyanonymous committed 1 year ago
suggest absolute full path to the `requirements.txt` instead of just `requirements.txt` (#7079)
2b14065 Dr.Lt.Data committed 1 year ago
Make it easier to set a custom template for hunyuan video.
65042f7 comfyanonymous committed 1 year ago
improved: better frontend package installation guide (#7047)
d6e5d48 Dr.Lt.Data committed 1 year ago
Use comfyui_frontend_package pypi package to manage frontend dependency (Frontend v1.10.17) (#7021)
04cf0cc Chenlei Hu committed 1 year ago
Rename argument in last commit and document the options.
4dc6709 comfyanonymous committed 1 year ago
--fast now takes a number as argument to indicate how fast you want it.
cf0b549 comfyanonymous committed 1 year ago
Use fp16 for intermediate for fp8 weights with --fast if supported.
eb45434 comfyanonymous committed 1 year ago
Use fp16 if checkpoint weights are fp16 and the model supports it.
1804397 comfyanonymous committed 1 year ago
Reduce artifacts on Wan by doing the patch embedding in fp32.
0270a0b comfyanonymous committed 1 year ago