ComfyUI Skills for OpenClaw

This project is a ComfyUI skill integration layer for OpenClaw, Codex, and Claude Code. It turns the workflows you build and export from ComfyUI in API format into callable skills that these agents can trigger with natural language.

It converts natural language requests into structured skill arguments, maps them to ComfyUI workflow inputs, submits jobs to ComfyUI, waits for completion, then pulls generated images back to local disk.

For the upstream ComfyUI local server routes that back this skill, see docs/comfyui-native-routes.md.

Features

Workflow-to-Skill Conversion

Turn your ComfyUI workflows into skills that OpenClaw, Codex, or Claude Code can call directly
Set up parameters in three steps: upload a workflow, pick the nodes to expose, then name and describe each parameter
Input nodes are detected automatically from the workflow — no need to hunt through the JSON by hand
When you update a workflow, preview the parameter changes first and carry over existing mappings
Supports bulk workflow import from a ComfyUI server or local JSON files

Multi-Server Management

Manage multiple ComfyUI servers and route jobs to different machines as needed
Servers and workflows each have their own toggle — agents only see a workflow when both are enabled
Workflows with the same name on different servers stay separate, identified by server/workflow composite IDs
Set a per-server output directory and choose a default server

Web UI

Frontend source lives in a separate repository; run scripts/update_frontend.sh to pull the latest build
Runtime update checks prefer pulling the main repo via git, and fall back to refreshing ui/static/ from the frontend release when git update is unavailable
A local web interface for managing all servers and workflows in one place
Reorder workflows by dragging, or sort by name, status, or custom order
Search and filter workflows across all servers
Available in English, Simplified Chinese, and Traditional Chinese — switch languages from the UI

Configuration Transfer

Export your current configuration and workflows into a single portable JSON file
Choose exactly which workflows to include before exporting
Preview what will change on the target machine before applying an import
Control whether to overwrite existing workflows or sync environment settings from the source machine

Workflow Execution

Submit a job, wait for it to finish, and download the generated images to local storage automatically
Supports multi-image output in a single run
Agents can query all available workflows and their parameters from the command line for automated use

CLI Tools

A full set of command-line tools for headless environments and scripting
Covers server management, workflow listing, job execution, and configuration import/export

ComfyUI Native API Scope

This repository should be understood as two layers:

Native ComfyUI server routes on the target generation server, such as /prompt, /history/{prompt_id}, /view, /ws, and /queue
This project's own manager API under /api/*, used by the local UI to manage saved servers, workflows, and transfer bundles

The current skill implementation is centered on the native execution flow:

POST /prompt
GET /history/{prompt_id}
GET /view

That route-level reference is documented in docs/comfyui-native-routes.md.

Installation

ComfyUI Skills for OpenClaw

Manual install:

cd ~/.openclaw/workspace/skills
git clone https://github.com/HuangYuChuh/ComfyUI_Skills_OpenClaw.git comfyui-skill-openclaw
cd comfyui-skill-openclaw
python3 -m venv .venv
source .venv/bin/activate
python3 -m pip install -r requirements.txt
cp config.example.json config.json

If python3 --version is below 3.10, create the virtual environment with any installed python3.10+ interpreter, for example python3.12 -m venv .venv.

Let OpenClaw install it for you:

Send this prompt to OpenClaw:

Please install this ComfyUI skill into my OpenClaw workspace.

Target path:
~/.openclaw/workspace/skills/comfyui-skill-openclaw/

Requirements:
1. Run `cd ~/.openclaw/workspace/skills` first.
2. Clone this repository into `comfyui-skill-openclaw`.
3. Keep SKILL.md at the project root.
4. Install Python dependencies from requirements.txt.
5. Run `cp config.example.json config.json`.
6. Set the default ComfyUI server URL to http://127.0.0.1:8188 unless I specify another one.
7. Make sure OpenClaw can discover and call this skill after installation.

ComfyUI Skills for Claude Code

Manual install:

cd ~/.claude/skills
git clone https://github.com/HuangYuChuh/ComfyUI_Skills_OpenClaw.git comfyui-skill
cd comfyui-skill
python3 -m venv .venv
source .venv/bin/activate
python3 -m pip install -r requirements.txt
cp config.example.json config.json

If python3 --version is below 3.10, create the virtual environment with any installed python3.10+ interpreter, for example python3.12 -m venv .venv.

Let Claude Code install it for you:

Send this prompt to Claude Code:

Please install this ComfyUI skill into my Claude Code skills directory.

Target path:
~/.claude/skills/comfyui-skill/

Requirements:
1. Run `cd ~/.claude/skills` first (create the directory if it doesn't exist).
2. Clone this repository into `comfyui-skill`.
3. Keep SKILL.md at the project root.
4. Install Python dependencies from requirements.txt.
5. Run `cp config.example.json config.json`.
6. Set the default ComfyUI server URL to http://127.0.0.1:8188 unless I specify another one.
7. Make sure Claude Code can discover and call this skill after installation.

ComfyUI Skills for Codex

Manual install:

cd ~/.codex/skills
git clone https://github.com/HuangYuChuh/ComfyUI_Skills_OpenClaw.git comfyui-skill
cd comfyui-skill
python3 -m venv .venv
source .venv/bin/activate
python3 -m pip install -r requirements.txt
cp config.example.json config.json

If python3 --version is below 3.10, create the virtual environment with any installed python3.10+ interpreter, for example python3.12 -m venv .venv.

Let Codex install it for you:

Send this prompt to Codex:

Please install this ComfyUI skill into my Codex skills directory.

Target path:
~/.codex/skills/comfyui-skill/

Requirements:
1. Run `cd ~/.codex/skills` first (create the directory if it doesn't exist).
2. Clone this repository into `comfyui-skill`.
3. Keep SKILL.md at the project root.
4. Install Python dependencies from requirements.txt.
5. Run `cp config.example.json config.json`.
6. Set the default ComfyUI server URL to http://127.0.0.1:8188 unless I specify another one.
7. Make sure Codex can discover and call this skill after installation.

Updating

To update the skill, run the update script and restart the UI or your agent runtime:

./update.sh

This pulls the latest code, syncs frontend assets, and installs any new Python dependencies in one step.

If you prefer updating manually, you can also run git pull followed by pip install -r requirements.txt.

If you use the update prompt in the Web UI, it follows the same repository-first update flow.

ComfyUI Workflow Setup

Before you start, make sure your ComfyUI server is already running. The default local address is http://127.0.0.1:8188.

UI Setup (Recommended)

macOS/Linux: ./ui/run_ui.sh, or double-click ui/run_ui.command
Windows: ui\run_ui.bat
Visit: http://localhost:18189
Upload a workflow JSON exported from ComfyUI with Save (API Format)
Add your first ComfyUI server in the UI
Select which parameters should be exposed to OpenClaw and save the mapping

Config File Setup

1) Edit `config.json`

Configure the server first. Minimal example:

{
  "servers": [
    {
      "id": "local",                  // Server ID, also used as the directory name and workflow prefix
      "name": "Local",                // Display name
      "url": "http://127.0.0.1:8188", // ComfyUI server URL
      "enabled": true,                // Whether this server is enabled
      "output_dir": "./outputs"       // Image output directory
    }
  ],
  "default_server": "local"           // Default server ID
}

2) Place Workflow Files

Each workflow uses its own directory, for example:

data/local/Default/
  workflow.json  # ComfyUI API-format workflow export
  schema.json    # Parameter mapping exposed to OpenClaw/Agent

3) Write `schema.json`

schema.json should include at least:

description
enabled
parameters

Minimal example:

{
  "description": "Default test workflow", // Human-readable description for OpenClaw/Agent
  "enabled": true,                        // Whether this workflow is enabled
  "parameters": {
    "prompt": {                           // Parameter name exposed to OpenClaw/Agent
      "node_id": 10,                      // Node ID in workflow.json
      "field": "prompt",                  // Input field name under that node
      "required": true,                   // Whether this field is required
      "type": "string",                   // Parameter type
      "description": "Prompt text"        // Parameter description
    },
    "seed": {
      "node_id": 10,
      "field": "seed",
      "required": false,
      "type": "int",
      "description": "Random seed"
    }
  }
}

Notes:

The workflow ID comes directly from the directory name. For example, if the directory is data/local/Default/, the workflow ID is Default
Each entry in parameters defines one input exposed to OpenClaw/Agent
node_id and field must match the actual node and input field in workflow.json

If you want a full example, refer to:

data/local/Default/workflow.json
data/local/Default/schema.json
These two files are generic examples. Before running them, replace node 4's ckpt_name in workflow.json with a checkpoint name that exists on your ComfyUI server.

4) Verify the Configuration

List the available workflows:

python scripts/registry.py list

Run a test generation:

python scripts/comfyui_client.py \
  --workflow <server_id>/<workflow_id> \
  --args '{"prompt":"test"}'

Example:

python scripts/comfyui_client.py \
  --workflow local/Default \
  --args '{"prompt":"A premium product photo"}'

On success, the output looks like this:

{
  "status": "success",
  "prompt_id": "...",
  "images": ["./outputs/<prompt_id>_...png"]
}

Agent-Assisted Setup

Let OpenClaw or another agent edit config.json
Let the agent write workflow.json and schema.json into the target workflow directory
After writing the files, let the agent run a verification step

Workflow Requirements

An API-format workflow plus a Save Image output node is the baseline requirement for stable use. To avoid failed or empty runs:

The workflow must be exported in ComfyUI API format
- In ComfyUI, click Save (API Format)
- Place the exported JSON at data/<server_id>/<workflow_id>/workflow.json
The workflow must end with a Save Image node
- The current client downloads results from ComfyUI output images
- Without a Save Image node (or equivalent image output), the workflow may finish but return no downloadable image

Multi-Server Management

You can configure multiple ComfyUI servers so OpenClaw can route jobs across different hardware targets such as a local GPU or a remote instance.

Core Concepts

Dual-Layer Toggles: Both servers and individual workflows can be enabled or disabled. A workflow is only visible to the AI agent if both the server and the workflow itself are enabled.
Namespacing: Workflows are identified with a composite ID: <server_id>/<workflow_id> (e.g., local/sdxl-base vs. remote-a100/sdxl-base).

CLI Configuration

On headless machines, you can use the built-in CLI tool scripts/server_manager.py:

python scripts/server_manager.py list
python scripts/server_manager.py add --id remote --name "Remote Node" --url http://10.0.0.1:8188
python scripts/server_manager.py disable remote

You can still manage all server settings through the web UI.

Configuration Migration

If you move this skill to another path or another machine, use the built-in bundle flow to transfer the current configuration and workflow mappings.

UI flow:

Click Export Config on the main page to download openclaw-skill-export.json
Before export, you can expand each server and uncheck workflows you do not want to include; all workflows are selected by default and servers are collapsed by default
Open the UI on the target machine and click Import Config
Select the exported JSON bundle
Review the preview summary, then decide whether to also apply the source machine's default server, URL, and output directory

CLI flow:

python scripts/transfer_manager.py export --output ./openclaw-skill-export.json
python scripts/transfer_manager.py import --input ./openclaw-skill-export.json --dry-run
python scripts/transfer_manager.py import --input ./openclaw-skill-export.json

Optional flags:

--apply-environment: also apply bundle default server, URL, and output directory during import
--no-overwrite: skip existing workflows instead of overwriting them

Default import behavior:

Existing workflows with the same ID are overwritten
Existing servers are merged instead of replaced
The target machine keeps its current url, output_dir, and default_server unless --apply-environment is used

Common Issues

If ComfyUI returns HTTP 400 on /prompt, the workflow payload or one of the parameter values is usually invalid.
size must match values accepted by the underlying node (e.g. 3:4,1728x2304).
If config.json points to the wrong server URL, job queueing will fail.

Changelog

See CHANGELOG.md for the full release history.

Roadmap

Workflow version history and rollback
Upgrade preview before applying a new workflow version
Parameter migration support when upgrading a workflow
Authentication support for remote ComfyUI servers
Better schema validation before queueing
Richer error reporting from ComfyUI node errors
Execution history with parameter and result tracking
Webhook callbacks on task completion
Scheduled workflow execution (cron-style)
Guided workflow rewrite recipes for agents
Native ComfyUI plugin with AI-assisted parameter mapping — describe what a workflow does in natural language inside ComfyUI, and let the AI automatically infer node-field relationships to generate schema.json, eliminating manual parameter configuration
AI agent workflow generation — describe your image/video generation goal in natural language, and let the AI agent automatically construct, connect, and configure a complete ComfyUI workflow graph from scratch

Project Structure

ComfyUI_Skills_OpenClaw/
├── SKILL.md                    # Agent instruction spec (how to call registry/client)
├── README.md
├── README.zh.md
├── LICENSE
├── .gitignore
├── requirements.txt            # Python deps (FastAPI, requests, etc.)
├── config.example.json         # Example runtime config
├── config.json                 # Actual local runtime config (gitignored)
├── asset/
│   └── banner-ui-dashboard-20260322.png
├── data/
│   ├── <server_id>/
│   │   └── <workflow_id>/
│   │       ├── workflow.json       # ComfyUI workflow API export
│   │       └── schema.json         # Exposed parameter mapping
├── scripts/
│   ├── server_manager.py       # CLI tool for managing servers
│   ├── registry.py             # List workflows + exposed parameters for agent
│   ├── comfyui_client.py       # Inject args, queue prompt, poll history, download images
│   └── shared/                 # Shared config & JSON utils (reused across scripts)
│       ├── config.py
│       ├── json_utils.py
│       └── runtime_config.py
├── ui/
│   ├── app.py                  # FastAPI app – routes only
│   ├── open_ui.py              # Agent-friendly UI launcher
│   ├── services.py             # Business logic (workflow CRUD)
│   ├── models.py               # Pydantic request/response models
│   ├── json_store.py           # Low-level JSON file read/write helpers
│   ├── settings.py             # App-level settings
│   ├── run_ui.sh               # Start UI (macOS/Linux)
│   ├── run_ui.command          # Double-click launcher (macOS)
│   ├── run_ui.bat              # Launcher (Windows)
│   └── static/                 # Modular ES6 frontend (HTML/CSS/JS)
└── outputs/
    └── .gitkeep

Project Keywords And Resources

Project Keywords

This repository is organized around the following search intents:

OpenClaw
ComfyUI
ComfyUI Skills
ComfyUI workflow automation
OpenClaw ComfyUI integration
AI image generation skill
Xiao Long Xia (small crawfish / 小龙虾, project nickname)

Core files for project understanding and retrieval:

README.md (English overview)
README.zh.md (Chinese overview)
SKILL.md (agent execution contract)
docs/llms.txt and docs/llms-full.txt (LLM-oriented summaries)

Project Resources

Project summary: docs/llms.txt
Extended project context: docs/llms-full.txt
Project discovery checklist: docs/PROJECT_DISCOVERY_CHECKLIST.md

ComfyUI Skills for OpenClaw

Features

Workflow-to-Skill Conversion

Multi-Server Management

Web UI

Configuration Transfer

Workflow Execution

CLI Tools

ComfyUI Native API Scope

Installation

Updating

ComfyUI Workflow Setup

UI Setup (Recommended)

Config File Setup

1) Edit config.json

2) Place Workflow Files

3) Write schema.json

4) Verify the Configuration

Agent-Assisted Setup

Workflow Requirements

Multi-Server Management

Core Concepts

CLI Configuration

Configuration Migration

Common Issues

Changelog

Roadmap

Project Structure

Project Keywords

Project Resources

Reviews (0)

1) Edit `config.json`

3) Write `schema.json`