qiaomu-codex-imagegen AI Skill

What it is

Qiaomu Codex Imagegen is a tool by GitHub user joeseesun. Codex's image generation normally only works inside Codex; this project opens it up through an MCP server, a CLI and an Agent Skill, so Claude Code, Cursor or any agent that can run commands can say "make a Xiaohongshu image" and get an image file back. Before each image, it suggests at least four style directions built on different mechanisms; once you pick, it writes the full prompt, generates, and the agent checks the result against acceptance criteria one by one. Codex in the name is OpenAI's product, and this repo is the author's own bridge to it. What sets it apart is that it proposes directions first instead of drawing right away.

Highlights:

24 scene templates with 48 presets, plus directions in the styles of 20 Mondo poster artists
preset ratios and safe areas for video covers, Xiaohongshu, WeChat headers, book covers and more
a JSON file beside each image records the actual prompt, so you can reproduce it or build a series
image editing from a reference picture.

How to use:

First make sure the Codex CLI is installed, signed in and able to generate images.
Then run npx skills add joeseesun/qiaomu-codex-imagegen, or register the MCP server as the README shows.
Finally ask your agent, for example "Make a 16:9 cover for this video."

Who it's for:

Content creators already on Codex who want covers and social images straight from other agents.

Before you start:

You need Node.js 18 or later and the Codex CLI installed and signed in, on an account that can generate images.
Images go through your Codex account, so prompts and reference images are processed by OpenAI under your plan's pricing and limits.
The author says the events, brands, people and copy in the samples are fictional, and small text in images may have flaws, so zoom in before using one.
MIT license.

Based on the project's public documentation. Not tested by AI Inspiration World.