ComfyUI
Node-based diffusion model GUI and API
109k|stable|v0.19.0
Graph/nodes workflow interface — Build complex Stable Diffusion pipelines without code using a visual node editor.
Multi-model support — Runs SD1.x, SDXL, SD3, Flux, Mochi, and 30+ other diffusion models.
Smart memory management — Automatically offloads models between VRAM and system memory, runs on 1GB GPUs.
Queue system with selective re-execution — Only re-computes graph nodes that changed between runs, skipping redundant calculations.
Workflow persistence — Save and load workflows as JSON, or embed them in PNG/WebP output files.
ComfyUI dominates the open-source diffusion UI space with a node-graph approach that lets you build production workflows without touching code. Weekly releases keep pace with new model drops (Flux 2, Hunyuan Video), and the smart memory management means you can run SDXL on consumer GPUs. Desktop apps for Windows/macOS make it accessible, but power users deploy the Python backend locally or on cloud instances.
GPU recommended (NVIDIA/AMD/Intel/Apple Silicon supported). Runs CPU-only with --cpu flag. Python 3.8+.
