What it does

Ian Xiaohei Illustrations is a Codex Skill that reads Chinese articles, identifies their “cognitive anchors” — judgments, flows, states, metaphors — and generates white-background, hand-drawn PNGs to drop between paragraphs. The default visual IP is “Xiaohei”: a solid black stick-figure with dot eyes, thin legs, and no expression, who must actively participate in whatever system the image depicts. Not a mascot. Not decoration. A deadpan actor in your metaphor.

The interesting bit

Most AI image prompts beg for “beautiful illustration.” This one begs for restraint: 40-60% whitespace, thin jittery black lines, minimal red/orange/blue Chinese annotations, and a rule that if removing Xiaohei doesn’t break the image, you’ve failed. The skill invents low-tech physical metaphors per article — “idea press,” “information well,” “trust bridge” — rather than recycling compositions from a template bank.

Key highlights

• Outputs a shot list (4-8 images) before generating, mapping each to a specific paragraph and cognitive anchor
• Enforces 16:9 horizontal format with strict style DNA: pure white, no gradients, no paper texture, no PPT-info-graph vibes
• Includes QA checklist to catch style drift, hallucinated labels, and decorative Xiaohei placement
• Saves PNGs to assets/<article-slug>-illustrations/ workspace path
• Ships with reference docs: style DNA, composition patterns, prompt templates, and IP guidelines

Caveats

• Chinese text in images is error-prone; shorter annotations are more stable, and bad typos require regenerating with fewer words
• Not for vector editing, PPT exports, or anyone wanting cute mascot stickers — the author explicitly warns away children-cartoon and brand-key-visual seekers
• Requires Codex/Claude Code environment; it’s a skill directory you copy into ~/.codex/skills/, not a standalone CLI tool

Verdict

Chinese content creators, newsletter writers, and methodology bloggers who want a recognizable, lightly absurd visual voice without hiring an illustrator. Skip it if you need editable diagrams, corporate polish, or your audience expects Dribbble-grade flat design.