ComfyUI now has an AI agent that builds workflows for you. Comfy Agent, available today on Comfy Cloud and coming to Comfy Desktop in a few weeks, can plan, construct, run, and iterate on image and video generation workflows based on text descriptions. The tool eliminates much of the manual node-wiring and technical debugging that slows down creative work.
"We strongly believe creators shouldn't have to choose between ambition and complexity. Comfy Agent is our answer to that," the Comfy team said. The agent sits inside the ComfyUI interface and understands the latest models, extensions, and workflows. It can evaluate techniques, fix errors, find assets, and translate technical requirements so users can focus on the creative idea.
How the agent works alongside you
Comfy Agent operates directly on your canvas while you edit. Both you and the agent can work without waiting for each other. You can reference another workflow, drag an asset from the library, or mention a node. The agent sees assets visually and notices when you move things on the graph. Every ComfyUI control remains accessible - you can open any node and check every parameter.
Users can create public or private skills while working, and run up to five chats in parallel with full chat history. The agent handles requests like updating a generation to compare three more video models, explaining what a workflow does and where it's slow, or re-running a product-shot workflow across 20 images.
What early creators built
Three creators tested the agent before launch. Animator 852芹 from Andidea used it to shape the look of an original character named YUI, explore scenes, and compare generation results over 100 hours of work. VFX artist Doug Hogan used the agent to isolate a subject, generate masks, and work with depth and normals in a single workflow that took a shot from image to compositing passes.
For professionals looking to build similar skills, ComfyUI Courses cover node-based workflows in depth. As AI agents become more capable of handling complex toolchains, AI Agent Courses can help teams understand how to integrate these assistants into production pipelines.
What's coming next
Local Comfy installations will soon become fully agentic, using your own GPU, models, and custom nodes. The agent will walk users through setup, pick models suited to their hardware, and customize the environment. The team is also developing agent-built custom nodes that will create tailored editing tools on demand. Additional features in the pipeline include more model options, improved public skills and knowledge, and broader external integrations.
Comfy Agent is free to try and uses existing Comfy Credits. Users can log in at cloud.comfy.org, open the Agent panel to claim free initial chats, and start with a real task like "Take this product photo and make 20 on-brand ad variants in 4 aspect ratios." The tool is in beta, and the team is collecting feedback through a dedicated form and the #comfy-agent channel on Discord.
Why this matters for creatives, marketers, and product teams
Comfy Agent reduces the gap between having a specific visual idea and executing it. For marketers producing ad variants, product developers iterating on visual concepts, or writers who need consistent image assets without learning node-based interfaces, the agent handles the technical plumbing. The time saved on debugging and workflow assembly shifts directly into creative iteration - trying models, adjusting composition, and comparing approaches until the result matches what was in your head.
Your membership also unlocks:
