Generate and run a ComfyUI workflow from a concrete request
At a glance
A Claude Code skill selects ComfyUI templates and node definitions to produce an importable workflow instead of requiring manual graph wiring. This workflow is listed under Workflow Generation, Image & Video, Developer Tools.
- Agents / tools
- Claude Code, ComfyUI, ComfyUI Workflow Skill, FLUX, Wan 2.2
SourceBuilding a tool for these workflows? See weekly rates.
What they did
1) Have Git, Claude Code and a working ComfyUI installation with hardware suitable for your chosen model. Run `mkdir -p ~/.claude/skills`, then `git clone https://github.com/LingyiChen-AI/comfyui-workflow-skill ~/.claude/skills/comfyui-workflow`. Check that SKILL.md exists there. 2) In Claude Code, describe the task and model family, such as generating a FLUX image followed by a Wan 2.2 image-to-video stage. The skill reads the relevant template and node specifications and saves a workflow JSON in the current directory. 3) Load that JSON in ComfyUI or drag it into the interface. Follow any missing-model download prompts and ensure the required models are available. 4) Click Queue to execute the graph, then inspect the generated image or video before changing the request or parameters for another run.
How it went
The discussion includes a FLUX-to-Wan demonstration and an output video. The linked repository also shows generated workflows being imported and run.
Worth knowing
Generating the graph does not supply GPU capacity. The repository notes that one motion-transfer demonstration used reduced resolution because of video-memory limits.
