Tool
ComfyUI
A modular, node-based interface and backend for diffusion models — build graphs for image (and related) generation on your own GPU instead of clicking a single “Generate” button.
What it is
ComfyUI (Comfy-Org/ComfyUI, historically comfyanonymous/ComfyUI) is a graph/nodes UI plus API for running Stable Diffusion, Flux, and related pipelines. You wire samplers, models, LoRAs, and ControlNets explicitly. Power users prefer it; beginners often bounce to simpler UIs first.
What it replaces
It replaces one-click web UIs when you need reproducible, complex pipelines and local control. Versus AUTOMATIC1111’s web UI, ComfyUI is more graph-native and less “all settings on one page.” It does not replace model licenses, commercial image APIs when you have no GPU, or taste.
License
The ComfyUI repository is licensed under the GNU General Public License v3.0 (GitHub license metadata). GPL-3.0 is strong copyleft for distributed modifications of the covered work. Custom nodes and model weights are separate legal objects — LoRAs and checkpoints have their own terms (CreativeML, Flux licenses, etc.).
Where it falls over
- VRAM is the product. 4 GB is pain; 12–24 GB is where modern pipelines breathe.
- Node graphs intimidate; broken custom nodes are a support genre.
- Model license compliance is on you — the UI will load whatever file you drop in.
- Video / large FLUX-class workflows need serious hardware or lots of patience.
Who it's for
People who already think in pipelines and have a GPU worth feeding. If you want the simplest local SD button, start with a simpler UI — then graduate here when you outgrow it.
Hardware that usually matters
The apps on these pages are free or free-to-use. The recurring cost is usually the machine: VRAM first, then system RAM, then disk for weights. Example Amazon listings for parts people actually buy for local inference — not endorsements; stock and price move constantly.
- ASUS Prime GeForce RTX 5060 Ti 16GB — mid-range 16 GB VRAM card; enough for many 7B–13B quantized models and comfortable SD/Comfy workflows.
- GIGABYTE GeForce RTX 3090 24GB — older generation, but 24 GB VRAM is still what a lot of people hunt for larger local models and heavier image pipelines.
- CORSAIR Vengeance DDR5 64GB (2×32GB) — system RAM matters when layers spill off the GPU or you run CPU offload.
- Samsung 990 PRO 2TB NVMe SSD — model files eat disk; a fast NVMe keeps pulls and swaps from feeling stuck.
Some links are Amazon Associates. We may earn a commission if you buy.
Related in this cluster
- Stable Diffusion WebUI — classic Gradio UI alternative
- whisper.cpp — local audio; different modality, same “own the box” idea
- AnythingLLM — local text/RAG instead of pixels
- Ollama — local LLMs on the same machine if VRAM allows
Verdict
Use ComfyUI when local, composable diffusion graphs are the job and your VRAM can pay the bill. Read model licenses separately from GPL on the app. Star counts do not render faster.