GitHub radar
ComfyUI: Visual Node Graph for Any AI Model — Image, Video, Audio, 3D (132k★)
A local node-graph editor that runs Stable Diffusion, Flux, Wan, LTX-Video, HunyuanVideo, and 30+ more models — wire any workflow you can imagine without writing code, with full control over every parameter.
ComfyUI is a visual node-graph editor for running AI generation models locally. You assemble pipelines by connecting blocks — a model loader, a text encoder, a sampler, a VAE decoder, a save node — and the data flow is visible before you hit run. It natively supports over 30 model families for image, video, audio, 3D, and text generation, including Stable Diffusion, Flux, Wan, LTX-Video, HunyuanVideo, and Qwen3. VRAM is managed automatically with smart offloading and partial graph re-execution, so changing one node does not restart the entire pipeline. Custom nodes extend the graph with thousands of community-built operations. Workflows are saved as JSON files, shareable and reusable. It runs fully offline — no telemetry, no mandatory cloud.
Why a vibe-coder should care
If you have hit the limits of A1111 or InvokeAI trying to chain ControlNet, LoRA, inpainting, and upscaling in one reproducible pipeline, ComfyUI handles it without plugins conflicting with each other. The graph format makes your workflow self-documenting: you can re-run it exactly a month later with the same settings, or share it so someone else gets the same result.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Go to https://github.com/comfyanonymous/ComfyUI — install ComfyUI following the README (desktop app or pip instructions for my OS), launch a text-to-image workflow from comfy.org/workflows, and show me the result. Ask me which model to pick if you need a starting point.
Requires any GPU with 4 GB+ of VRAM — NVIDIA, AMD, Intel, or Apple Silicon; CPU mode is available if you don't have a GPU, just slower.
Open on GitHub▌ More finds