Installation guide
freestylefly/awesome-gpt-image-2 installation guide
MIT-licensed GPT-Image2 prompt catalog, gallery, website, and installable Agent Skill that lets Claude Code select repository-backed templates, styles, scene tags, pitfalls, and example cases to produce structured image-generation prompts.
Install `awesome-gpt-image-2` for Claude Code on macOS Tahoe (Apple silicon)
What you are installing
`freestylefly/awesome-gpt-image-2` is primarily a Prompt-as-Code gallery and template library for GPT-Image2. It contains 544 documented example cases, more than 20 prompt templates, and a generated style reference covering UI, infographics, posters, products, branding, architecture, photography, illustration, characters, scenes, historical themes, and publishing.
For Claude Code, the relevant component is the **GPT-Image2 Style Library Agent Skill**. It helps Claude select a repository-backed template, style tags, scene tags, constraints, and nearby example cases, then produce a copyable image-generation prompt. Installing the skill does **not** itself provide an image-generation API or automatically generate images.
Claude Code classification
- **Mechanism:** `context_injection`
- **Install mode:** `claude_extension`
- **Setup effort:** low
This classification is supported by the repository’s `SKILL.md` package and its documented Claude Code skill location. Claude Code receives the skill’s Markdown workflow and reference library as task context. The supplied files do not establish an MCP server, lifecycle hooks, command interception, or a separate executable image-generation tool, so those mechanisms should not be attributed to this repository.
Before you begin
This guide is for macOS Tahoe on an Apple-silicon Mac. The supplied repository documentation does not identify any Intel-only components or architecture-specific setup. The easiest Claude Code method uses Claude Code’s own plugin marketplace commands and avoids cloning or building the website.
You need:
- A working Claude Code installation.
- Internet access so Claude Code can add the GitHub-hosted marketplace and install the skill.
- For the alternative `npx` or npm methods only, a working Node.js/npm environment. The repository does not document a Node installation procedure, so this guide does not provide one.
You do **not** need Supabase, Vercel, Stripe, Google Analytics, Alipay, or a GPT Image API key merely to install and use the Claude Code skill. Those services belong to the repository’s separately documented website and hosted-generation setup.
Recommended installation: Claude Code plugin marketplace
Step 1: Open Claude Code
Start a Claude Code session in the normal way you already use it. The next two entries are Claude Code slash commands, not macOS Terminal shell commands.
Step 2: Add the repository as a marketplace
Paste this into Claude Code and submit it:
/plugin marketplace add freestylefly/awesome-gpt-image-2This registers the repository’s documented Claude Code marketplace source.
Step 3: Install the style-library plugin
After the marketplace command finishes, paste:
/plugin install gpt-image-2-style-library@awesome-gpt-image-2This installs the named GPT-Image2 Style Library plugin from that marketplace. No Apple-silicon-specific flags are documented or required by the supplied files.
Step 4: Restart the session
The repository tells users to restart the agent session after installing. End the current Claude Code session and start a fresh one using your normal workflow. There is no source-supported restart command to paste.
Verify the installation
In the new Claude Code session, paste the repository’s English example request:
Use gpt-image-2-style-library to create an infographic prompt about Codex.A successful response should use the installed skill to select an appropriate template and produce a copyable GPT-Image2 prompt. According to `SKILL.md`, the expected workflow identifies the output category, matches a template and style or scene tags, and includes concrete constraints such as text, layout, aspect ratio, and negative details. It should also identify the selected template and useful example-case IDs when applicable.
This verifies skill use, not image generation. You must take the resulting prompt to a compatible image-generation service if you want an actual image.
Alternative: install with the `skills` CLI
Use this route if you prefer Terminal or want the same skill installed for both Claude Code and Codex. Open macOS Terminal, paste the following single documented command, and press Return:
npx skills add freestylefly/awesome-gpt-image-2 --skill gpt-image-2-style-library --agent claude-code codex --global --yes --copyThis selects only `gpt-image-2-style-library`, targets Claude Code and Codex, installs globally, accepts prompts automatically, and requests copied files. The repository recommends this method for Claude Code, Codex, Cursor, and other tools supported by the `skills` package.
If you deliberately want installation into every local agent supported by that installer, the repository instead documents:
npx skills add freestylefly/awesome-gpt-image-2 --global --all --copyDo not use the all-agent command merely for Claude Code unless you actually want files placed for other supported agents too. After either command finishes, restart your Claude Code session and use the verification request shown above.
Alternative: use the published npm CLI
The repository publishes a dedicated package named `gpt-image-2-style-library`. The simplest documented npm form runs it without globally installing its CLI:
npx gpt-image-2-style-library install allBe aware that `install all` is broader than a Claude-Code-only installation. The documentation says it writes the skill into common local folders including:
- `~/.claude/skills`
- `~/.codex/skills`
- `~/.agents/skills`
The inspected installer confirms that it creates the target skill directory, removes an existing `gpt-image-2-style-library` directory at that target, and copies `SKILL.md`, `agents`, `assets`, and `references` into place. Therefore, preserve any personal modifications inside an existing installed copy before rerunning it.
If you prefer a globally installed CLI, the repository documents these commands:
npm install -g gpt-image-2-style-library
gpt-image-2-style-library install allRestart Claude Code afterward. The package also accepts a Claude-Code-only target according to the supplied installer’s usage text:
gpt-image-2-style-library install claude-codeUse that last command only after the documented global CLI installation above. It targets Claude Code’s skill folder rather than all three agent locations.
Which method should you choose?
For most laypeople, use the **Claude Code plugin marketplace** method. It is the shortest Claude-native path and requires only two slash commands inside Claude Code.
Use the **`skills` CLI** if you already have `npx` and want a copied skill shared with Codex or other supported agents. Use the **published npm CLI** if you specifically prefer npm-managed installation or need its explicit Claude-Code-only target.
Choose one primary method; installing the same skill through several methods is unnecessary and can make it unclear which copy Claude Code is reading.
Using the skill effectively
Ask Claude Code explicitly to use `gpt-image-2-style-library`, then describe the image you want. The skill is designed for creating, rewriting, classifying, or improving GPT-Image2 prompts. It can route requests across categories such as posters, product images, realistic photography, character sheets, infographics, UI screenshots, and historical scrolls.
If your request is vague, the skill’s documented behavior is to offer two or three promising directions and ask you to choose. For a precise result, include the subject, exact visible wording, desired aspect ratio, layout priorities, and anything the image must avoid.
The repository’s disclaimer matters: many cases were organized from publicly accessible community material, and the project does not guarantee that third-party content is cleared for commercial use. Review source links and obtain permission from the relevant rights holder before commercial reuse. The repository itself is listed under the MIT License, but that does not automatically replace the rights or terms attached to third-party examples.