The free Codex skill that turns any concept into hand-drawn explainer images

You have probably seen them by now: clean, white-background sketches where a little black blob character cranks a machine, pulls a lever, or stands inside a diagram acting out an idea. The golden ratio, AIDA, the 4Ps, network effects — each one drawn like a page from a smart, slightly absurd notebook. They look hand-crafted and intentional, and the surprising part is that they come from one free tool you feed a plain description to. It is called Xiaohei, and here is exactly what it is and how to run it yourself.

Xiaohei — the name literally means “little black” — is a Codex skill built by the developer helloianneo. A skill is just a folder of instructions that Codex loads and follows, so instead of prompting an image model from scratch every time, you hand Codex your concept and it already knows the house style: pure white background, black hand-drawn linework with a little wobble, generous white space, and a few sparse handwritten annotations in red, orange, or blue. The blob is not decoration. The skill is built so the character has to physically do the thing the idea is about — turn the crank, hold the two halves apart, ride the boat into the blue ocean. Remove the character and the picture stops making sense, which is what makes each one feel like an explanation rather than a stock illustration.

Normally, getting a visual like this means a designer and an afternoon in Figma, or settling for a generic slide template that looks like everyone else's. Xiaohei collapses that into a sentence of input and one core idea per image.

Installing it

You need Codex installed. From there, the setup is four lines in your terminal — you clone the repo and copy the skill into your Codex skills folder:

git clone https://github.com/helloianneo/ian-xiaohei-illustrations.git

cd ian-xiaohei-illustrations

mkdir -p "${CODEX_HOME:-$HOME/.codex}/skills"

cp -R ./ian-xiaohei-illustrations "${CODEX_HOME:-$HOME/.codex}/skills/"

That drops the skill into the folder Codex reads on startup. Once it's there, you just describe what you want in plain language and Codex picks up the skill automatically.

The one setting everyone misses: language

This is the part worth flagging up front. Xiaohei was written for Chinese article writers, so by default the annotations in the images come out in Chinese. If you want English labels — the “why special?”, “remove a square”, “value innovation” text you see in the examples — you have to say so. Add “in English” to your prompt, or tell it plainly to put all handwritten labels in English. Skip this and you'll get a lovely diagram annotated in a language your team may not read.

Three ways to prompt it

Plan first, generate later. Paste in a whole article and ask it not to draw yet — just to tell you which parts are worth illustrating and return a shot list of roughly five images. This is the smart opening move for a long post: you see the visual plan before spending any generations, then pick the ones you actually want.

Generate straight from an article. Drop in the text and ask for, say, four Xiaohei body illustrations in English. It reads the piece, finds the key judgments and metaphors, and draws them.

One concept at a time. Give it a single idea or one sentence — “the whole relates to the large part as the large part relates to the small” — and ask for one image. This is how you get a clean, single-purpose diagram for a specific term or framework, which is exactly what those marketing-concept images are.

Getting good results

A few habits make the difference between a keeper and a reroll. Keep the text you want on the image short — long labels are where these tools garble letters, and short phrases render far more reliably. Hold each image to one core idea; if you're trying to cram a whole framework into a single frame, split it into a few. And review every output before you ship it. Because it's drawing text, you'll occasionally catch a typo, a hallucinated label, or a bit of style drift — a quick pass fixes it, and if the words keep coming out wrong, ask it to use fewer annotations.

Where this really earns its place is exactly the use case in the reel: explaining ideas to a team. Marketing frameworks, onboarding flows, product concepts, mental models — anything that's normally a wall of bullet points becomes a single image someone actually remembers. One description in, one intentional-looking diagram out, and the little black blob does the explaining for you.

The repo is here: github.com/helloianneo/ian-xiaohei-illustrations. Clone it, remember to ask for English, and start with a shot list.