Communitygithub.com

Raw Takes to Published Video: honest high-CTR thumbnail

Generates several high-CTR thumbnail concepts for a finished video: layout and focal subject, an overlay of four words or fewer, color and contrast logic, the emotional read, and a ready image-generation prompt for each. Includes an honesty check so the thumbnail matches the real content.

What is Raw Takes to Published Video: honest high-CTR thumbnail?

Raw Takes to Published Video recipe, step 3 (package it), from Starlight-Intelligence/starlight-agent-skills (MIT). thumbnail-concept takes the video topic and its promise and returns a JSON list of concepts (default three), each with layout, overlay text of four words or fewer, color logic, emotional read and an image-gen prompt. Its quality bar demands legibility at small size, one clear focal point, no clickbait, and a rights note when real agency imagery is used. Works with any image-generation tool or MCP.

Works with~Claude Code~Codex CLI~Cursor
npx skills add https://github.com/Starlight-Intelligence/starlight-agent-skills/tree/HEAD/skills/media/thumbnail-concept

Ask in your favorite AI

Open a new chat with this agent skill pre-loaded.

Documentation

Thumbnail Concept

A video topic → several high-CTR thumbnail concepts with image-gen prompts.

Purpose

The thumbnail decides whether good content gets watched. This skill produces a few distinct, honest, high-contrast thumbnail concepts — each with a layout, a short text overlay, a color logic, and a ready image-generation prompt — so the creator can pick and generate quickly.

When it fires

  • Keyword triggers: thumbnail, cover art, thumbnail ideas, ctr, video cover
  • File triggers: thumbnails/**
  • Command triggers: /thumbnail-concept

Inputs

InputRequiredNotes
topicyesVideo subject + the promise.
subjectnoHero image/asset (e.g. a nebula, a rocket).
countnoNumber of concepts, default 3.

Workflow

  1. Find the visual hook — the single image idea that conveys the promise. ✅ Check: readable at small size.
  2. Per concept: layout (subject placement, rule-of-thirds), overlay text (≤4 words, high contrast), color/contrast logic, and emotional read.
  3. Honesty check — the thumbnail must match the actual content (no bait).
  4. Image-gen prompt — a concrete prompt to produce the background/subject; if using NASA imagery, note the rights step.

Output contract

{
  "concepts": [{
    "layout": "string",
    "overlay_text": "string — ≤4 words",
    "color_logic": "string",
    "emotional_read": "string",
    "image_prompt": "string"
  }]
}

Tools & MCP

  • Tools: Read. MCP: optional image-gen MCP.

Quality bar

  • Overlay text ≤4 words and legible at thumbnail size.
  • Concept matches the real content (no clickbait).
  • High contrast / clear focal point.
  • Rights noted if real agency imagery is used.

Example

Input: "Why Webb sees what Hubble can't" → 3 concepts, each with a 2–3 word overlay, contrast logic, and an image prompt.


Built on SIP — Starlight Intelligence Protocol Substrate: starlightintelligence.org/protocol v1.1.0 Layers used: [file-contract, attestation, sovereignty] Vertical: starlight-agent-skills · portable capability layer

Individual skills in this repo

This repo contains 1 individual skill — each has its own dedicated page.

Related Skills