ComfyUI

You download the desktop app for your own hardware or sign up for Comfy Cloud, then load a community template or build a graph from scratch. Nodes represent models, processing steps and outputs, and you connect them on the canvas so each decision is explicit: which checkpoint, which sampler, how many steps, what conditioning, what post-processing. Running the graph executes the pipeline step by step, and because every stage is a node you can change one parameter and re-run without rebuilding anything. Comfy Agent sits inside ComfyUI and will plan, build and run a workflow from a plain-language description, editing the canvas alongside you rather than replacing it. When a workflow is finished, the Developer Platform deploys it as an API your own application can call, and the Comfy API runs workflows and models from your code. On Cloud, you are charged for active GPU time while a workflow runs; idle time spent building does not consume credits.

• •Node Graph Control - Models, conditioning, samplers and outputs wired explicitly on an infinite canvas, with every parameter visible and adjustable rather than hidden behind a prompt
• •Self-Hosted and Open Source - GPL-3.0, runs on your own GPU through the desktop build, with no account and no per-image cost
• •Comfy Agent - Describe a result and the agent plans, builds and runs the workflow inside ComfyUI, editing the canvas while you work
• •Deploy as an API - The Developer Platform turns a finished workflow into a production endpoint, with builds, deployments and workers per plan
• •Community Node Ecosystem - More than 60,000 nodes and thousands of shared workflows, with Python custom nodes for anything missing
• •Comfy Cloud - The same workflows on hosted Blackwell RTX 6000 Pro GPUs with 96GB VRAM and 900+ pre-installed models, billed on active GPU time
• •Partner Nodes - Proprietary models run through the same credit pool, usable on both Cloud and local installs

People doing repeatable visual production work: VFX and animation, advertising and creative studios, games, and ecommerce or fashion imagery, where the same pipeline runs many times and a prompt box does not give enough control. Developers get the clearest value from the Python custom node API and the deploy-as-an-API path, which turn a graph into infrastructure something else can call. Researchers and model authors use it to inspect and modify a sampling pipeline directly. Be honest about the fit, though: most ComfyUI users are producing images and video, not shipping software, so if you came looking for a coding tool this is not one. The node graph has a real learning curve and is the wrong choice if you want a result in one sentence, where a hosted text-to-image service is simpler. And if you intend to embed it in a closed-source commercial product, read GPL-3.0 first.