gpt-image-2
Teach your coding agent to generate and edit images with gpt-image-2: prompt grammar, exact text rendering (Chinese included), and a zero-dependency CLI for generation, edits, and inpainting.
- What
- Teach your coding agent to generate and edit images with gpt-image-2: prompt grammar, exact text rendering (Chinese included), and a zero-dependency CLI for generation, edits, and inpainting.
- Cost
- Free
- Needs
- Use "gpt-image-2" (from the claude-image repo) with a coding agent that has shell access (e.g. Claude Code).
- Install
- Copy the installer prompt below into your Muse β your agent does the rest.
π³ Paid API required. Curated by Skill Harbor: this is the skill from the claude-image repo, and it teaches a coding agent to generate, edit, and iterate on images with the gpt-image-2 model. It covers the prompt grammar (open with intent, quote every text string exactly, one style anchor, no magic words), a zero-dependency CLI (scripts/gpt_image.py) that handles auth, retries, parallel batching, and file IO, precise local edits via a change-ONLY-X / preserve-Y-exactly pattern, mask-based inpainting, a size table per use case, and a strict visual self-verification loop: the agent reads each PNG itself and checks text rendering, composition, and style before showing it to you. By @jiangmuran, listed here with credit to its creator. Discovered via github.com/topics/ai-skills. Honest caveats: needs a paid API key for gpt-image-2; the default API host in the repo is a third-party proxy (jmrai.net), you can override it with your own OpenAI-compatible host; the host does not support multi-image input or transparent backgrounds; this skill is built for coding agents with shell access in the Claude Code style, not for chat-only assistants; the repo includes a few Chinese examples. Skill Harbor never reviews the code, review it yourself before use.
Version:
Install
Copy the install package below, then paste it into MuseCommunity-built. Skill Harbor doesn't audit code β review the source before installing.
Use "gpt-image-2" (from the claude-image repo) with a coding agent that has shell access (e.g. Claude Code). Prerequisites: Python 3, a paid API key for the gpt-image-2 model. The repo defaults to a third-party proxy host; you may use your own OpenAI-compatible host instead. 1. Fetch the repo: https://github.com/jiangmuran/claude-image, then copy SKILL.md, the scripts/ folder, and the references/ folder into a temporary folder. 2. Safety check: read SKILL.md and scripts/gpt_image.py for anything suspicious (the script must only talk to the image API host you configured; no credential harvesting). Your API key goes in an environment variable only, never in the code or the chat. STOP on any red flag and tell me. 3. Install it as a skill: copy SKILL.md, scripts/ and references/ into the agent's skills directory, in a folder named "gpt-image-2". 4. Export your key on your machine: OPENAI_IMAGE_API_KEY, and optionally OPENAI_IMAGE_BASE_URL if you use your own host (never paste the key into the chat). 5. Verify: run python3 scripts/gpt_image.py --help from the skill folder. Then generate one small test image and read the PNG yourself to confirm text renders correctly. 6. Report what was installed, where, and what I still need to do myself (e.g. saving my API key). Tip: for edits, describe ONLY the change plus what to preserve exactly ("change ONLY X, preserve Y exactly"). Pick the size from the skill's table before writing the prompt; a wrong aspect ratio wastes generations. Safety: every generation is billed. Keep the key in an environment variable or the secure vault, review each command before it runs, and visually verify every result before sharing it.
Saved to your recent installs. Find it anytime on /connect.
Questions
How do I install a build?
Every product page includes a copy-paste install prompt. Paste it into your Muse and it sets the build up for you β no manual configuration.
Where does my money go?
Straight to the seller. Skill Harbor never processes payments: checkout happens on the sellerβs own page, usually Stripe.
What does the β next to a creatorβs name mean?
It means we confirmed the identity of the person behind the listing. It says nothing about the code itself β always check a build before installing it.