gpt-image-2-skill

GPT-Image-2 prompt skill + MCP server for Claude Code and Codex.
Describe your intent → get a production-ready prompt → generate the image, all in one conversation.

Synthesized from three community repos:

• YouMind-OpenLab/awesome-gpt-image-2 — 2000+ curated prompts

• ZeroLu/awesome-gpt-image — top creator picks from X/Twitter

• freestylefly/awesome-gpt-image-2 — 470+ reverse-engineered cases, 20+ industrial templates

Requirements

• Node.js ≥ 18

• Claude Code (CLI or desktop)

• OpenAI API key with Images API access

Installation

Step 1 — Run the installer

npx gpt-image-2-combined install

This one command will:

• Copy the Skill to ~/.claude/skills/gpt-image-2-combined/

• Register the MCP server in ~/.claude.json

• Create ~/.gpt-image-2/.env for your API key

Step 2 — Add your OpenAI API key

Open ~/.gpt-image-2/.env and fill in your key:

# ~/.gpt-image-2/.env
OPENAI_API_KEY=sk-...

Don't have a key? Get one at platform.openai.com/api-keys.

Step 3 — Restart Claude Code

The Skill and MCP server load on startup. Restart once after installation.

That's it. Verify the Skill is available by typing /gpt-image-2-combined in Claude Code.

Usage

Full flow (prompt → image in one conversation)

1. Invoke the Skill to craft a prompt

/gpt-image-2-combined 幫我做一張章魚的 Apple keynote 風格科普海報

The Skill will:

• Detect your intent and language

• Match the best template (poster-nature-science in this case)

• Fill in the 6-block structure (subject, composition, style, text, format, constraints)

• Output a production-ready, copyable prompt

2. Generate the image via MCP

Once you have the prompt, ask Claude to generate it:

用這個 prompt 生成圖片

Claude calls the generate_image MCP tool, which:

• Sends the prompt to OpenAI Images API (GPT-Image-2)

• Downloads the result

• Saves it to ~/Pictures/gpt-image-2/

• Returns the local file path and CDN URL

3. Done

✅ Image generated and saved.

File : ~/Pictures/gpt-image-2/img-2025-06-19T10-30-00.png
URL  : https://...
Size : 1024x1536   Quality: standard

Skill only (prompt engineering without image generation)

If you only need a well-crafted prompt to paste into ChatGPT or another tool:

/gpt-image-2-combined make a sports campaign poster for a running shoe brand

Copy the output prompt and use it anywhere.

Template categories

MCP tool reference

The MCP server exposes one tool: generate_image.

Configuration

All config lives in ~/.gpt-image-2/.env:

# Required
OPENAI_API_KEY=sk-...

# Optional — verify exact model name at platform.openai.com/docs/models
GPT_IMAGE_MODEL=gpt-image-alpha

# Optional — where generated images are saved
GPT_IMAGE_OUTPUT_DIR=/Users/you/Pictures/gpt-image-2

What's inside

gpt-image-2-skill/
├── agents/skills/gpt-image-2-combined/
│   ├── SKILL.md                 ← Claude Skill (prompt engine)
│   ├── agents/
│   │   ├── openai.yaml          ← Codex agent spec
│   │   └── claude.yaml          ← Claude agent spec
│   └── references/
│       └── combined-tricks.md   ← All templates and tricks
├── mcp/
│   └── src/index.js             ← MCP server (calls OpenAI API)
├── scripts/
│   ├── cli.mjs                  ← npx entry point
│   └── install.mjs              ← install logic
├── .claude-plugin/
│   └── marketplace.json         ← Claude Code marketplace registration
├── .github/workflows/
│   ├── publish.yml              ← auto npm publish on git tag
│   └── test.yml                 ← cross-platform install test
└── .env.example                 ← API key template

Manual installation (alternative)

If you prefer to clone and install locally:

git clone https://github.com/albert850904/gpt-image-2-skill.git
cd gpt-image-2-skill
npm install
npm run install:skill

Releasing a new version

# bump version in package.json, then:
git tag v1.0.1 && git push --tags

GitHub Actions publishes to npm automatically on every tag push.

Credits

License

MIT