Generate more controllable article-cover directions with an AI Skill

The baoyu-cover-image Skill is the article-cover workflow inside JimLiu/baoyu-skills. Instead of relying on one vague prompt, it organizes a cover direction across six controllable dimensions: Type, Palette, Rendering, Text, Mood, and Font.

Input, result, and first use

Give it a Markdown article and it can choose or apply a cover direction. Type includes hero, conceptual, typography, metaphor, scene, and minimal; Palette ranges from warm and cool to mono and duotone; Rendering includes flat-vector, hand-drawn, painterly, digital, pixel, chalk, and screen-print. Text can be omitted, title-only, title plus subtitle, or denser. Mood is subtle, balanced, or bold. Font can be clean, handwritten, serif, or display.

The basic invocation is /baoyu-cover-image path/to/article.md. The default ratio is 16:9, with 2.35:1, 4:3, 3:2, 1:1, and 3:4 also supported; --no-title avoids asking the image model to render a title. For a first test, generate one non-sensitive article cover automatically, record its six decisions, hold five fixed while changing one, then compare title and no-title outputs.

Provider and text boundary

The upstream requires Node.js and an environment that can run npx bun. It recommends installing only needed Skills because the larger repository contains more than 20 Skills and can burden Agent context. Image generation still depends on a configured provider, including OpenAI, Google, Azure OpenAI, OpenRouter, DashScope, MiniMax, Replicate, Jimeng, or Seedream. Quality, cost, speed, and data handling depend on that provider.

For Traditional Chinese text, keep typography as a separate post-production step where possible. Image models can misspell Chinese, especially long titles, branded type, and small text. Important covers still need human selection and finishing; six dimensions do not automatically solve composition or brand consistency. Check rights and platform policies when prompts or output include people, trademarks, or a close imitation of a creator's style.

Source and license