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

Add a model merge node for SD3.5 large.
comfyanonymous committed 2 years ago
5281090
support SimpleTuner lycoris lora for SD3 (#5340)
PsychoLogicAu committed 2 years ago
af8cf79
Fix ControlLora issue with last commit.
comfyanonymous committed 2 years ago
66b0961
Clean up some controlnet code.
comfyanonymous committed 2 years ago
754597c
Fix lowvram edge case.
comfyanonymous committed 2 years ago
915fdb5
remove attention abstraction (#5324)
contentis committed 2 years ago
5a8a489
Optimizations to --fast and scaled fp8.
comfyanonymous committed 2 years ago
8ce2a10
Fix duplicate sigmas on beta scheduler.
comfyanonymous committed 2 years ago
f82314f
Mixed precision diffusion models with scaled fp8.
comfyanonymous committed 2 years ago
0075c6d
Support scaled fp8 t5xxl model.
comfyanonymous committed 2 years ago
83ca891
Fixed model merging issue with scaled fp8.
comfyanonymous committed 2 years ago
f9f9faf
fp8 casting is fast on GPUs that support fp8 compute.
comfyanonymous committed 2 years ago
471cd3e
Support diffusion models with scaled fp8 weights.
comfyanonymous committed 2 years ago
a68bbaf
Clamp output when rounding weight to prevent Nan.
comfyanonymous committed 2 years ago
73e3a9e
Add tooltips to LoraSave node.
comfyanonymous committed 2 years ago
518c0dc
ce0542e
Pytorch can be shipped with numpy 2 now.
comfyanonymous committed 2 years ago
8473019
Use the lowvram cast_to function for everything.
comfyanonymous committed 2 years ago
6715899
Add missing import.
comfyanonymous committed 2 years ago
7390ff3
Revert "Fix Transformers FutureWarning (#5140)"
comfyanonymous committed 2 years ago
0bedfb2
Add an experimental node to sharpen latents.
comfyanonymous committed 2 years ago
f71cfd2
c695c4a
Add some latent operation nodes.
comfyanonymous committed 2 years ago
0dbba9f
Cleanup some useless lines.
comfyanonymous committed 2 years ago
f584758
Fix Transformers FutureWarning (#5140)
svdc committed 2 years ago
95b7cf9
Switch default packaging workflows to python 3.12
comfyanonymous committed 2 years ago
191a0d5
Fix fp8 ops staying enabled.
comfyanonymous committed 2 years ago
3c60ecd
Remove useless argument.
comfyanonymous committed 2 years ago
7ae6626
model_options consistency between functions.
comfyanonymous committed 2 years ago
6632365