Coverbean

Generate ready-to-post cover / thumbnail images for 小红书 · 微信公众号 · 抖音 · TikTok · YouTube — from inside your AI agent.

Coverbean is a Claude Agent Skill. You describe the post; your agent writes the title and the image prompt; Coverbean renders it and crops it to the exact size each platform expects. Powered by apimodels.app's gpt-image-2-all.

• Five platforms, exact sizes — 3:4, 2.35:1, 9:16, 16:9, cropped precisely (via ffmpeg).
• Great text-in-image — short titles render crisp and legible.
• Cheap — a 1K cover is $0.005, so $1 of credit is ~200 covers.
• No lock-in on thinking — the agent (Claude, etc.) does the copywriting for free; you only pay for the render.

Platform	Output size
小红书 (xiaohongshu)	1080 × 1440 (3:4)
微信公众号 (wechat)	1080 × 459 (2.35:1)
抖音 (douyin)	1080 × 1920 (9:16)
TikTok (tiktok)	1080 × 1920 (9:16)
YouTube (youtube)	1280 × 720 (16:9)
all five (all)	—

Setup (about a minute)

1. Sign up at https://apimodels.app and add a little credit (a cover is $0.005 — $1 ≈ 200 covers).
2. Create an API key at https://apimodels.app/console/api-keys.
3. Export it:

   export APIMODELS_API_KEY=sk-xxxxxxxx

Also needs Node.js 18+, and optionally ffmpeg (for exact cropping).

Install

Claude Code — one-line plugin install:

/plugin marketplace add stimQQ/coverbean-skill
/plugin install coverbean@coverbean

Then just ask Claude: "Make me a YouTube thumbnail about 'AI is coming for your job'."

Manual (any agent) — copy the skill into your skills dir:

git clone https://github.com/stimQQ/coverbean-skill /tmp/coverbean
cp -r /tmp/coverbean/skills/coverbean-skill ~/.claude/skills/coverbean-skill

Or run the bundled script directly:

APIMODELS_API_KEY=$APIMODELS_API_KEY node skills/coverbean-skill/scripts/cover.mjs \
  --platform youtube \
  --prompt 'Bold thumbnail: wide-eyed creator pointing at a glowing AI chip, dramatic lighting, saturated teal/orange, large bold headline "AI 抢工作?" across the top third, high contrast' \
  --out ./covers

Use --platform all to render all five at once. Defaults: gpt-image-2-all, --quality low, --resolution 1K.

How it works

你: 给这篇文章做个 YouTube 封面
  └─ 你的 Agent(Claude)  →  起标题 + 写图像提示词   (免费,不花你钱)
       └─ Coverbean (cover.mjs)  →  gpt-image-2-all 出图 → 下载 → ffmpeg 精裁到平台尺寸
            └─ ./covers/youtube-cover.png  (1280×720, ready to upload)

The prompt recipe (short quoted title, one bold subject, high contrast, per-platform vibe) lives in SKILL.md so the agent writes covers that actually convert.

Notes

• Generated files on apimodels.app expire after 7 days — the script downloads them locally right away.
• Estimate cost at https://apimodels.app/tools; browse all models at https://apimodels.app/models.

---

Powered by apimodels.app — one key, many models (image · video · LLM · audio).