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
CLIPTextEncodeSDXL now works when prompts are of very different sizes.
2c9d98f comfyanonymous committed 3 years ago
Add arguments to run the VAE in fp16 or bf16 for testing.
e7bee85 comfyanonymous committed 3 years ago
Fix loras not working when loading checkpoint with config.
603f02d comfyanonymous committed 3 years ago
Add gpu variations of the sde samplers that are less deterministic but faster.
e57cba4 comfyanonymous committed 3 years ago
Add logit scale parameter so it's present when saving the checkpoint.
f81b192 comfyanonymous committed 3 years ago
Now the model merge blocks node will use the longest match.
c02f3ba comfyanonymous committed 3 years ago
ConditioningAverage now also averages the pooled output.
3a09fac comfyanonymous committed 3 years ago
Add text encode nodes to control the extra parameters in SDXL.
d94ddd8 comfyanonymous committed 3 years ago
Switch to fp16 on some cards when the model is too big.
2c4e0b4 comfyanonymous committed 3 years ago
Try to keep text encoders loaded and patched to increase speed.
b6a60fa comfyanonymous committed 3 years ago
Make highvram and normalvram shift the text encoders to vram and back.
97ee230 comfyanonymous committed 3 years ago