Xiaohei Article Illustration

Problem

Chinese technical articles often pack complex decisions, process turns, and state changes into long paragraphs, forcing readers to reread for structure. Generic illustrations usually fall into three traps: they look like PPT infographics, commercial artwork, or cute decoration. Xiaohei Article Illustration is not about producing a picture book. It selects cognitive anchors from the article and uses a 16:9 hand-drawn visual to clarify one core relationship, making the image part of the explanation rather than a stack of elements.

Workflow and boundaries

The skill uses Xiaohei as the default visual subject: a solid black figure with white dot eyes, thin legs, and a neutral expression. Xiaohei must perform the core action in the frame, not stand aside as decoration. The workflow typically has four steps: read the article, link, or Markdown to extract the core claim, cognitive turns, and places that benefit from an image; produce a shot list specifying where each image belongs, its theme, structure type, what Xiaohei does, and suggested Chinese labels; when generation is explicitly requested, call the built-in image_gen for each image separately, without combining layouts or copying prior examples; then check against references/qa-checklist.md for issues such as too many elements, PPT-like composition, wrong Chinese labels, or dirty background, and decide whether to regenerate or edit.

It fits Chinese articles, technical blogs, product notes, and process retrospectives that need a small number of inline illustrations, especially for input-output loops, breakpoints, branching, before-after comparisons, and common pitfalls. It is not intended for batch posters, commercial KV, or complex system architecture diagrams. For long articles, do not easily exceed 9 images; for short ones, stay within 1 to 3, otherwise the text is diluted by illustrations.