What’s inside

One SKILL.md (the playbook) + one starter template. A preview:

Social Cover Images

When to use

• The user is publishing a WeChat official-account post and needs a cover image (often asks "what size, and where does the title go?").
• The user is posting to Xiaohongshu/Douyin and needs a 3:4 portrait cover, worried the title will be hidden behind the system UI.
• The user is posting to Weibo/X/Jike and needs a 16:9 or 1:1 image with a "click-worthy" look.
• The user runs a course/event/podcast and wants a consistent set of covers (define one spec, reuse it).
• The user already has a base image and needs the headline overlaid with safe margins to make a usable cover.

Workflow

Ask where it will be published first, then set the canvas: WeChat post cover 2.35:1 (e.g. 900×383), Xiaohongshu 3:4 portrait (e.g. 1080×1440), Weibo/X 16:9 (e.g. 1200×675), square 1:1 (e.g. 1080×1080). Do one platform per image — no "one image for every platform."

Main headline ≤ 12 Chinese characters (one-sentence hook; suspense is fine, clickbait is not), sub-headline ≤ 20 characters (say what it's actually about), plus an optional garnish tag (e.g. "Episode 3", "field-tested"). If the copy can't be locked down, get the user to decide the words first — the image comes second.

Pattern A, centered: main headline as big centered type, generous whitespace, best for opinion/insight pieces; Pattern B, split columns: text left / image right (or vice versa), best for product/person introductions; Pattern C, stacked layers: image on the top 1/3, colour block + text on the bottom 2/3, best for events/courses. Mark the safe margins in the plan: keep 8–10% of each edge free of key text, and for Xiaohongshu leave 12% top and bottom (against the top bar and bottom caption overlay).

Emotion first, then colours: celebratory/promo → reds/oranges; professional insight → deep blue/ink; fresh lifestyle → off-white/pale green. One primary + one accent + one text colour is enough — white text on dark backgrounds, dark grey/black text on light ones. Contrast guaranteed.

→ output: complete prompt) Write the prompt in English (image models handle English more reliably). It must include: canvas ratio, composition pattern, colour hex values, and a

1. Lock platform and size (input: publishing platform → action: look up the spec table → output: confirmed canvas spec)
2. Split the headline into three levels (input: article topic/title → action: three-level split → output: three-level copy)
3. Pick a composition pattern (input: three-level copy → action: choose one of three → output: composition plan)
4. Set the colour mood (input: content emotion → action: pick colours → output: hex palette)
5. Write the image-gen prompt (input: spec + copy + composition + palette → action: fill the template.md skeleton

Download the full skill pack (ZIP)