HuangYuChuh/ComfyUI_Skills_OpenClaw

What it solves

ComfyUI Skills for OpenClaw provides a bridge that allows AI agents to execute ComfyUI workflows without needing to manipulate complex, raw JSON graphs. It transforms these workflows into simple, callable "skills" with a controlled interface, ensuring agents can trigger image or video generation tasks safely and reliably through a CLI.

How it works

The project implements a schema-based parameter mapping layer. Users import ComfyUI workflows (in API format) and define a schema.json that maps specific agent-friendly aliases to the actual node IDs and fields within the ComfyUI graph. This allows an agent to send a simple command like prompt: "a white cat" instead of modifying a deep JSON structure.

It provides a CLI (comfyui-skill-cli) for execution and management, and an optional Web UI for visual configuration. It also integrates with the official local Comfy MCP (Model Context Protocol) for advanced discovery of nodes and models.

Who it’s for

• AI Agent Developers: Those using OpenClaw, Hermes Agent, Codex, or Claude Code who want their agents to have generative AI capabilities via ComfyUI.
• ComfyUI Power Users: Users who wanting to reuse their existing complex workflows as tools for other software or agents.
• Multi-Server Admins: Users managing multiple local or remote ComfyUI instances across different machines.

Highlights

• Agent-friendly CLI: A dedicated interface designed for reliable agent interaction rather than human use.
• Schema-based Mapping: Exposes only necessary parameters to the agent, hiding the complexity of the workflow graph.
• Multi-server Routing: Manage and route jobs to multiple local or remote ComfyUI servers under a single namespace.
• Dependency Management: Ability to check for missing nodes or models before executing a workflow.
• Visual Configuration: An optional Web UI to upload workflows and visually map parameters before deploying them to agents.