leon-llb/codex-image

Codex Image is a small Python helper that lets command-line tools and AI agents generate images for free by routing the request through your existing ChatGPT Plus subscription. The author's pitch is that you do not need an OpenAI API key, you do not pay per image, and you do not go through a third-party proxy. Instead it piggybacks on a separate macOS desktop app called Codex.app, which ships its own codex CLI that already talks to ChatGPT. A comparison table sets out the trade-off. With Codex Image the cost is included in your Plus plan, you get the ChatGPT native image model with its text-rendering quality, and setup is a single git clone. The trade-off is that this only works on macOS, because Codex.app is macOS-only, and you still need an active ChatGPT Plus subscription. Usage is direct. After cloning the repo into your skills directory, you run python3 generate.py with a prompt string, an optional size like 1024x1024, and an optional output directory. The README gives three example prompts for a square shiba inu illustration, a portrait cinematic shot, and a landscape cyberpunk wallpaper. Files are saved with a timestamped filename like codex-image-20260518-143052.png. The project also documents two integrations. The first is Claude Code, Anthropic's coding assistant: dropping the skill into the skills directory makes it discoverable, and saying generate an image of in conversation triggers it. The second is Hermes, also called OpenClaw Agent, which is a Telegram and WeChat chat agent. There are step-by-step instructions for copying a plugin file into the Hermes config, enabling the plugin in config.yaml, and restarting the gateway so chat messages can request images. The How it works diagram traces the flow from user, to Hermes or Claude, to an image_generate tool call, to the codex-image provider, to generate.py, to a codex exec call against ChatGPT Plus, and finally an image saved to ~/Downloads. The project is MIT licensed.