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

Increase maximum batch size in LatentRebatch.
comfyanonymous committed 3 years ago
ee74ef5
Refactor the template manager (#1878)
Matteo Spinelli committed 3 years ago
6e84a01
d2e27b4
Allow model or clip to be None in load_lora_for_models.
comfyanonymous committed 3 years ago
2455aae
Not used anymore.
comfyanonymous committed 3 years ago
e73ec8c
Fix some issues with sampling precision.
comfyanonymous committed 3 years ago
111f1b5
7c0f255
Remove a bunch of useless code.
comfyanonymous committed 3 years ago
a268a57
Sampling code changes.
comfyanonymous committed 3 years ago
1777b54
Added Bayer dithering to Quantize node.
tsone committed 3 years ago
23c5d17
Fix some memory issues in sub quad attention.
comfyanonymous committed 3 years ago
c837a17
Fix some OOM issues with split attention.
comfyanonymous committed 3 years ago
125b03e
41b07ff
Add --max-upload-size argument, the default is 100MB.
comfyanonymous committed 3 years ago
a12cc05
Cleanup webp import code a bit.
comfyanonymous committed 3 years ago
aac8fc9
Fix checkpoint loader with config.
comfyanonymous committed 3 years ago
2a134bf
Restrict loading embeddings from embedding folders.
comfyanonymous committed 3 years ago
434ce25
40963b5
Faster clip image processing.
comfyanonymous committed 3 years ago
723847f
Fix some OOM issues with split and sub quad attention.
comfyanonymous committed 3 years ago
a373367
Fix uni_pc returning noisy image when steps <= 3
comfyanonymous committed 3 years ago
7fbb217
change 'c_adm' to 'y' in ControlNet.get_control
Jedrzej Kosinski committed 3 years ago
3783cb8
Pass extra conds directly to unet.
comfyanonymous committed 3 years ago
d1d2fea
036f88c