Comfy Agent builds and runs ComfyUI workflows from a sentence

1 week ago

Comfy Agent, from Comfy Org, plans, builds and runs ComfyUI workflows from a plain-language request, on the same canvas you work on. Announced 1 October 2026; live on Comfy Cloud in beta, free to try with Comfy Credits, and coming to Comfy Desktop in a few weeks. Launch post: https://blog.comfy.org/p/comfy-agent-the-first-agent-for-craft Try it: https://cloud.comfy.org Videos: the launch post's own, including Doug Hogan's VFX use case and the YUI animation by 852話 and Andidea, cut short and muted.

Ask

Ask about this presentation

Answers are generated from this presentation.

Chapters

  1. 0:00Comfy Agent builds and runs ComfyUI workflows from a sentence
  2. 0:15Ask for your artwork on new product shapes, and it builds the graph
  3. 0:30Pick a favourite, and it cuts a 10-second ad
  4. 0:39A VFX artist turns a performer in a grey suit into a photoreal Bigfoot
  5. 0:55Then the layers a compositor needs: a matte, depth and normals
  6. 1:05One artist, 100 hours: an animated short about an original character, Yui
  7. 1:17Free to try on Comfy Cloud; next, the agent comes to the app on your own computer
  8. 1:33Comfy Agent is on Comfy Cloud now, and comes to Comfy Desktop in a few weeks
Show transcript

Comfy Agent builds and runs ComfyUI workflows from a sentence

On October first, Comfy Org released Comfy Agent, an assistant that lives inside ComfyUI, the node-based app many artists use to make AI images and video. You describe what you want, and it plans the workflow, wires up the nodes, and runs it on the canvas in front of you.

Ask for your artwork on new product shapes, and it builds the graph

Here's a session from Comfy's own launch video. The request: put our product artwork on some different product shapes, and make a video of it. The agent lays out the nodes and runs them. Then it's asked for four blank jars in different shapes and materials, to use as references, and it makes those too.

Pick a favourite, and it cuts a 10-second ad

Choosing is plain English too. Here the request is two more camera angles of the stylish jar, cut together into a ten-second ad. This is the ad it made.

A VFX artist turns a performer in a grey suit into a photoreal Bigfoot

Doug Hogan, a VFX artist, tried it on a real shot. He gives the agent a performer in a grey suit and asks for a photoreal Bigfoot walking through a forest, using his own restyling workflow with MiniMax's H3 model. On the left, the plate he started from. On the right, the result.

Then the layers a compositor needs: a matte, depth and normals

Then he asks for the comp passes, and gets them: a matte made with SAM three, a depth pass, and a normals pass. Those are the layers a compositor uses to fit a creature into a shot.

One artist, 100 hours: an animated short about an original character, Yui

Three creators had it before launch. One artist spent a hundred hours with the agent on this animated short about an original character called Yui, using it to shape the look, explore scenes and compare what different models made.

Free to try on Comfy Cloud; next, the agent comes to the app on your own computer

It's live now for everyone on Comfy Cloud, in beta. It's free to try, and it spends the Comfy Credits you already have. Next, Comfy says the local app will become fully agentic: your own graphics card, your own models and custom nodes, with the agent setting it up and picking models that fit your hardware.

Comfy Agent is on Comfy Cloud now, and comes to Comfy Desktop in a few weeks

Announced 1 October 2026 by Comfy Org, in a post by Jo Zhang, Christian Byrne and Kishore Shimikeri. In beta.

The launch postblog.comfy.org
Try itcloud.comfy.org

Comfy Org announced it on October first, 2026. It's on Comfy Cloud now, and Comfy says it comes to Comfy Desktop in a few weeks. The launch post and the cloud app are linked below.