This project provides a "skill" for AI coding agents like Claude Code and Cursor, enabling users to generate ComfyUI workflow JSON files directly from natural language descriptions. It targets users who want to leverage ComfyUI for complex generative AI tasks (image, video, audio, 3D) without the steep learning curve of manually connecting nodes, offering a significant productivity boost for rapid prototyping and workflow creation.

How It Works

The skill translates natural language prompts into ComfyUI's JSON workflow format. It utilizes 34 built-in templates and over 360 node definitions extracted directly from ComfyUI's source code to ensure accuracy. A key feature is the automatic model download capability; workflows include a models field that prompts ComfyUI to download missing models upon import, supporting sources like HuggingFace and CivitAI. This approach abstracts away the complexity of ComfyUI's node editor, allowing for intuitive, conversational workflow design.

Quick Start & Requirements

• Prerequisites: Claude Code installed, Git installed.
• Installation:
  1. Clone the repository: git clone https://github.com/twwch/comfyui-workflow-skill.git ~/codes/comfyui-workflow-skill
  2. Create a symbolic link to the Claude Code skills directory: ln -s ~/codes/comfyui-workflow-skill ~/.claude/skills/comfyui-workflow
  • Alternative methods include direct cloning or manual copying.
• Verification: Run ls ~/.claude/skills/comfyui-workflow/SKILL.md.
• Optional LLM Integration: Requires installing the comfyui_LLM_party custom node in ComfyUI (cd ComfyUI/custom_nodes; git clone https://github.com/heshengtao/comfyui_LLM_party.git; pip install -r comfyui_LLM_party/requirements.txt).

Highlighted Details

• Supports 34 built-in workflow templates for various models and tasks.
• Includes definitions for over 360 ComfyUI nodes, ensuring field accuracy.
• Automated model downloading prompts for missing diffusion models, LoRAs, etc.
• Supports a wide range of generative models: SD1.5, SDXL, SD3, FLUX, Wan2.2, HunyuanVideo, LTXV, Mochi, Cosmos.
• Integrates with LLM providers (OpenAI, Claude, Gemini, Ollama, etc.) via the comfyui_LLM_party custom node for advanced workflows like script-to-video.

Maintenance & Community

No specific details regarding maintainers, sponsorships, or community channels (e.g., Discord, Slack) are provided in the README. The project structure focuses on templates and node definitions.

Licensing & Compatibility

• License: MIT.
• Compatibility: The MIT license permits commercial use and integration into closed-source projects.

Limitations & Caveats

Users must have ComfyUI installed and configured. If using LLM integration or other advanced features, corresponding custom nodes (e.g., comfyui_LLM_party) must be installed separately in ComfyUI. Generated JSON values must strictly adhere to ComfyUI's widget type and range constraints to avoid runtime errors.