Comfy Agent: Describe Your Workflow, Watch It Build
Your ComfyUI workflows just got a co-pilot. On October 1st, ComfyUI launched Comfy Agent, an AI assistant that lives inside the node editor and builds, edits, and debugs workflows without leaving the canvas. Instead of describing what you want and getting back a JSON file to import, you watch the agent place nodes, wire them, and iterate—while the graph stays edible and visible the whole time.
This matters because ComfyUI's power is also its steepness. You can chain models, LoRAs, upscalers, and video nodes into sophisticated pipelines, but finding the right combination, getting the syntax right, and debugging when a node breaks kills iteration speed. Comfy Agent flattens that curve.
What people are building
VFX artist Doug Hogan is using it to isolate subjects, generate masks, and work with depth and normals in one workflow—taking a shot from image to compositing passes. Animation workflows are rolling out too: one creator used it to shape the look, explore scenes, and compare generation results for an original character YUI story. The canonical example is product photography: ask Comfy Agent to "take this product photo and make 20 on-brand ad variants in 4 aspect ratios", and it builds the pipeline.
The agent doesn't just execute once. It runs generations with your permission, reviews the output, and in auto mode it runs and iterates on its own until the result is ready.
How it works, and what it can't do
Describe what you want and Comfy Agent plans the workflow, drops the nodes onto the canvas, wires them together, runs the graph and iterates when something breaks. It works on the canvas you already have open rather than handing back a JSON file to import, so the graph stays visible and editable while it builds.
It can explain and debug any workflow by reading the entire graph, pointing at slow parts and tracing errors. It also carries curated knowledge for model recommendations and node selection.
The real constraint isn't the agent's intelligence—it's the nodes available. Local Comfy will soon be fully agentic, and the agent will walk you through GPU setup, pick models that fit your hardware, and customize your environment. Right now it's on Comfy Cloud with Desktop coming in a few weeks.
How to try it
Comfy Agent is open to everyone on Comfy Cloud from day one, with a version for Comfy Desktop promised in a few weeks. It's free to try and uses the Comfy Credits you already have. The feature is in beta.
Start at cloud.comfy.org, log in, open the "Agent" panel, and claim your free initial chats. The setup is transparent: you keep the workflow, you can hand-edit anything the agent builds, and every node stays visible as it assembles. Up to five chats retain separate histories and context, so you can iterate on parallel ideas without losing your train of thought.