Skip to content

Note gpt-image2-skill

All posts
Nov 15, 20267 min readgpt-image2-skill · ai-image-generation · claude-code · open-source · agent-skills · openaiRead in:ENID

GPT-Image2-Skill: Generate AI Images From Your Terminal With 171 Prompts

GPT-Image2-Skill launched in April 2026 and already has 6,100 stars on GitHub. A 171-prompt gallery, two agent skills, and a CLI bring OpenAI image generation into your terminal.

A code editor screen with an AI Actions menu open. Photo by Daniil Komov on Pexels.

You have a good idea for an image, but the loop is slow: open a web app, type a prompt, wait, download, repeat. GPT-Image2-Skill moves that whole loop into the terminal. It is a free, open source CLI and agent skill library for OpenAI image models that gives coding agents like Claude Code a gallery of 171 curated prompts, so you can generate, edit, and reverse engineer images without leaving your coding workflow. It has already passed 6,100 stars on GitHub. (GitHub)

This post explains what GPT-Image2-Skill is, how its two skills work, what the 171 prompts cover, and how to install it.

Key Takeaways

GPT-Image2-Skill is a free, open source CLI and agent skill library from wuyoscar, MIT licensed, with 6,100 stars and 489 forks on GitHub as of November 15, 2026.

It ships a gallery of 171 numbered prompts across 31 categories, from portraits and product shots to pixel art and infographics.

Two skills are included: gpt-image generates and edits images, and get-prompt-from-image reverse engineers a reference image into a reusable prompt.

It works with Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, and other agents that support the open Agent Skills format, or as a standalone CLI.

Image generation defaults to GPT Image 2.6 Flare at 1024x1024, high quality, and you pay OpenAI for the API calls.

What Is GPT-Image2-Skill?

GPT-Image2-Skill is a prompt gallery, an agent skill, and a CLI in one package. The project describes itself as a "GPT Image 2/2.5/2.6 prompt gallery, image prompt library, agentic skill, and CLI for OpenAI image generation/editing". In plain terms, it turns a coding agent into a design assistant: you describe the image you want in words, the agent picks a matching prompt template from the gallery, calls the OpenAI Images API, and saves the result to your project folder. (GitHub)

It was created on April 22, 2026, and it keeps pace with OpenAI's model releases: the current default is GPT Image 2.6 Flare, the smaller model built for speed, while Sunburst is the base model for detail or precise edits. Both support generation, editing, and transparent backgrounds. (OpenAI)

Watch the 30-second repo signal on Repomaniac to see it in action.

The 171-Prompt Gallery

The heart of the project is its gallery: 171 numbered prompts across 31 categories, updated November 9, 2026. Instead of guessing how to phrase a request, you browse the gallery and pick a template that already works.

Aspect

Detail

Size

171 numbered prompts across 31 categories

Coverage

Portraits, product shots, landscapes, pixel art, infographics, logos, and more

Language

README available in English and Chinese

Format

Numbered entries, so an agent can reference prompt 42 instead of pasting a wall of text

Updates

Gallery grows with the README, last updated November 9, 2026

The numbering matters more than it looks. When an agent skill loads, it does not need to read all 171 prompts up front. The Agent Skills format uses progressive disclosure: at startup the agent reads only each skill's name and description, about 50 to 100 tokens each, and loads the full instructions only when a task matches. The prompt library works the same way, with the agent pulling the specific template it needs. (Skillsboard)

Two Skills, Two Jobs

The project ships two agent skills, and they solve opposite problems.

Aspect

Detail

gpt-image

Generates and edits images from text, using the prompt gallery and the packaged CLI

get-prompt-from-image

Reverse engineers a reference image into a reusable prompt

The gpt-image skill is the main one. It routes text to image requests to the Images API generations endpoint, reference based edits to the edits endpoint, and inpainting to the edits endpoint with a mask. It is a runbook, not a reimplementation: it tells the agent how to use the prompt library and the CLI, and it must not rewrite the image API code itself. (SkillMD)

The get-prompt-from-image skill, contributed by @LunarXuan, works in reverse. Give it an image you like, and it writes a positive prompt in Chinese (450 to 700 characters) plus an English version, and a negative prompt of 10 to 15 English phrases. That is useful when you want to reproduce a style you found elsewhere: instead of guessing the wording, you get a prompt you can reuse and tweak. (MCPInstall)

How to Install GPT-Image2-Skill

Installation depends on which tool you use. Python 3.11 or newer is required for the CLI.

For Claude Code, add the marketplace and install the plugin:

/plugin marketplace add wuyoscar/gpt_image_2_skill

/plugin install gpt-image@wuyoscar-skills

For Codex, add both skills with one command:

npx --yes skills@latest add wuyoscar/gpt_image_2_skill --skill gpt-image get-prompt-from-image --agent codex --copy

For the standalone CLI, no agent needed:

uvx --from git+https://github.com/wuyoscar/gpt_image_2_skill gpt-image

The CLI reads your `OPENAI_API_KEY` from the process environment, then from `.env`, then from `~/.env`. Successful API calls bill your OpenAI account, so keep an eye on usage. The defaults are sensible: model gpt-image-2.6-flare, size 1024x1024, high quality, and you can generate 1 to 10 images per call in png, jpeg, webp, or gif. (GitHub)

What You Can Do With It

The practical difference is that images become part of the build, not a separate chore. A few workflows that fit naturally:

  • Generate a hero image or social card for a project without opening a browser, then commit it next to the code.

  • Edit a screenshot by describing the change in words, for example removing an element or changing the background.

  • Take a reference image from a design you like, reverse engineer its prompt, and generate variations in the same style.

  • Batch generate assets for a game or a demo, with each image saved straight into the project folder.

The OpenAI side supports custom sizes in multiples of 16, with aspect ratios from 1:3 to 3:1, and quality options from auto up to max. Prompts can be up to 32,000 characters long, which is far more than any gallery template needs. (OpenAI)

GPT-Image2-Skill vs a Web Image Generator

The models are the same, so the difference is the workflow, not the pixels.

Aspect

Web image generator

GPT-Image2-Skill

Where you work

A browser tab

Your terminal and coding agent

Prompt help

You write it from scratch

171 curated templates to start from

Editing

Upload and retype

Reference edits and inpainting from the CLI

Reverse prompts

Not built in

get-prompt-from-image writes the prompt for you

Output

Downloads folder

Saved straight into your project

Cost

Subscription or credits

You pay OpenAI per API call

That said, the CLI is not a free lunch. You need an OpenAI API key, and every image bills your account. If you only generate a few images a month, a web app with a flat subscription may be cheaper. The skill earns its keep when images are part of a daily coding workflow, where staying in the terminal saves more time than the API calls cost.

Why It Took Off

GPT-Image2-Skill launched in April 2026 and reached 6,100 stars by November 15, with 489 forks and 3 releases. Part of the appeal is timing. Agent skills became an open standard this year, developed by Anthropic, and developers are collecting skills the way they used to collect dotfiles. A skill that gives any coding agent a working image studio is an easy install.

The other part is the gallery itself. Good image prompts are hard to write, and the difference between a vague request and a precise one is the difference between a usable asset and a throwaway. A library of 171 tested prompts, plus a skill that can reverse engineer any image you like, removes the hardest part of the job: knowing what to ask for.

Frequently Asked Questions

Is GPT-Image2-Skill really free?

Yes. The project is open source under the MIT license, and the CLI and skills cost nothing. You still need an OpenAI API key, and image generation bills your OpenAI account per call.

Do I need a coding agent to use it?

No. The standalone CLI works on its own through `uvx`, with no agent involved. The skills are for Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Windsurf, Cline, OpenCode, and other agents that support the open Agent Skills format.

What is the difference between Flare and Sunburst?

GPT Image 2.6 Flare is the smaller model, built for speed, and it is the default in GPT-Image2-Skill. Sunburst is the base model, built for detail or precise edits. Both support generation, editing, and transparent backgrounds.

Can it edit an existing image?

Yes. Reference based edits and inpainting with a mask go through the Images API edits endpoint, so you can change part of an image or restyle the whole thing from a description.

Does it work with other image models?

No. The CLI and skills are built around OpenAI's GPT Image models. Note that gpt-image-1 was shut down on October 23, 2026, so the current generation is 2.5 and 2.6.

Conclusion

GPT-Image2-Skill does one thing well: it puts image generation where developers already work. A gallery of 171 prompts, a skill that knows how to use them, and a reverse prompt skill for the images you admire, all inside the terminal. It is free, open source, actively maintained, and it works with the coding agent you already use.

If you build things and keep wishing you had a designer on call, install it and generate one image for your current project. The first asset it drops into your folder will tell you more than any article can.

Baca versi Bahasa Indonesia dari artikel ini di GPT-Image2-Skill: Buat Gambar AI dari Terminal dengan 171 Prompt.

Tags: gpt-image2-skill · ai-image-generation · claude-code · open-source · agent-skills · openai

All posts