Description
Generating images with diffusion models through a panel full of switches makes swapping a sampler or adding ControlNet a scavenger hunt. ComfyUI draws the whole pipeline as a node graph where every step is a node you wire up: model loading, prompts, sampling, upscaling, post processing, with workflows you can save and share. It also covers video, audio and 3D models. GPL-3.0, runs locally, with a desktop installer.
Node workflows: wire visually, keep complex pipelines readable, save them as files.
Broad model support: Stable Diffusion, Flux and other image models, plus video and audio models.
Low VRAM tuning: automatic memory management to run big models on smaller GPUs.
Recompute only what changed: editing a node re-runs just the affected part.
API and extensions: an API and a huge custom node ecosystem.
Features
Node workflows: wire visually, keep complex pipelines readable, save them as files.
Broad model support: Stable Diffusion, Flux and other image models, plus video and audio models.
Low VRAM tuning: automatic memory management to run big models on smaller GPUs.
Recompute only what changed: editing a node re-runs just the affected part.
API and extensions: an API and a huge custom node ecosystem.
