GitHub radar
A Codex Skill Draws Article Illustrations With One Recurring Character (Xiaohei)
Ian Xiaohei Illustrations is a Codex Skill - the agent reads your article, finds the main points in it, and draws one picture for each, always with the same character.
The author calls it a Codex Skill - an instruction file that teaches an AI agent to make illustrations for Chinese articles, posts, blogs, Notion pages and methodology texts. The agent first reads the text and looks for key moments in it - a key judgment, a turn of thought, a before-and-after comparison, a metaphor. From this it builds a plan for four to eight pictures (one for each such moment) and draws them one by one with the image generation tool built into Codex. Every picture follows the same recipe - horizontal 16:9 format, clean white background, black hand-drawn line, a bit of red, orange or blue handwritten captions in Chinese. Each scene has the same character - Xiaohei, a black silhouette with two white dots for eyes and thin legs, and by the skill's own rules he has to be doing something in the scene (not just standing next to it). The author writes directly what it's not good for - commercial illustration, PPT infographics or cartoon icons, the skill doesn't fit there.
Why a vibe-coder should care
The value here is not in the pictures themselves, but in the fact that the visual style is locked into one skill file and repeats the same way on every run - instead of a random picture each time, you get a recognizable signature for the content. In its current form the skill works only with Chinese text and Chinese captions on the pictures, so it doesn't fit directly for Russian or English articles. But the technique itself - give the agent a fixed character, one style and rules for picking moments (all in one file) - carries over to any language. There is nothing to lose by trying it - the license is MIT, and the pictures are drawn by the image generation tool already built into Codex, no need to set up a separate model.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Set up this Codex skill from https://github.com/helloianneo/ian-xiaohei-illustrations — clone the repo and copy the ian-xiaohei-illustrations folder into ~/.codex/skills per the README, then run one of the ready-made examples from examples/prompts.md and show me the resulting shot list, the images, and the path they were saved to.
A regular laptop is more than enough — there is no heavy computation here, just skill files for the agent. Images are drawn by Codex's own built-in generator, so you don't need to set up a separate model.
Open on GitHub▌ More finds