GPT Image — AI skill for Claude Code

GPT Image

A cross-runtime image skill for Claude Code and GitHub Copilot CLI that uses the bundled GPT Image CLI as the default path, with an explicit Codex mode when the user asks for it.

[中文说明 / Chinese README](./README_ZH.md)

Features

• CLI-first generation - Use the bundled scripts/image_gen.py by default for generation and editing
• Explicit Codex mode - Use Codex's built-in image workflow only when the user explicitly wants Codex/native generation
• Image editing - Support direct edits, reference-guided generation, and compositing flows
• Batch generation in CLI mode - Use generate-batch for many prompts in the default CLI workflow
• Shared prompting system - Reuse the same prompt structure across both modes

Two Modes

Mode	When to use	Backend	Needs OPENAI_API_KEY
Explicit CLI (default)	Normal generation/edit requests unless the user explicitly asks for Codex/native generation	scripts/image_gen.py	Yes
Codex mode	User explicitly asks for Codex, native generation, or Codex CLI	codex exec -> Codex built-in image workflow	No

Prerequisites

Explicit CLI mode

1. Python environment with the bundled script available
2. OpenAI Python package
3. OPENAI_API_KEY exported locally

Install dependencies if needed:

pip install openai # or use uv pip install openai
pip install pillow  # optional, for downscaling only

Codex mode

1. Codex CLI installed via npm i -g @openai/codex
2. Codex authenticated via codex login
3. This skill available to Claude/Copilot in the runtime's custom skill path

Quick checks:

codex --help
codex exec --help

Installation

Claude Code

Copy this directory into your Claude skills directory:

mkdir -p ~/.claude/skills/gpt-image
cp -R /path/to/gpt-image/. ~/.claude/skills/gpt-image/

Copilot / project-local skill setup

Keep the `gpt-image` folder in the location your Co