Video Thumbnail Generator

SkillMedia

Create platform-ready thumbnails from real video frames. Use when the user wants a thumbnail, cover image, YouTube cover, Shorts cover, Bilibili cover, Xiaohongshu cover, or other video poster image.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the Video Thumbnail Generator skill

What this skill tells your AI

The instructions your AI receives, as published by 0xsline/openchatcut in src/agent/skills/video-thumbnail-generator/SKILL.md and read by ahel’s review.

Use this skill to create a high-clarity thumbnail or cover image from the actual video content.

Non-Negotiables

  1. Use the real video as visual evidence.

    • If the thumbnail shows a person, product, UI, frame, or scene from the video, extract or inspect real frames.
    • Never invent a generic software interface, fake product screen, or unrelated person.
  2. Read style references before recommending styles.

    • Use references/thumbnail-styles.md and references/thumbnail-style-recipes.md.
    • Only offer style directions supported by those references or by the user's explicit instruction.
  3. Keep text short and readable.

    • Usually 2-5 words.
    • Avoid thin type, low contrast, small labels, and clutter.

Workflow

  1. Confirm platform and aspect ratio if missing.
    • YouTube / Bilibili: usually 16:9.
    • Shorts / TikTok / Douyin: usually 9:16.
    • Xiaohongshu / WeChat Channels: usually 3:4.
    • Correct ambiguous ratio language gently, especially 4:3 vs 3:4.
  2. Read project state, transcript, and available media context.
  3. Inspect candidate frames:
    • expressive face
    • product or UI clearly visible
    • dramatic action moment
    • clean composition
    • strong lighting or clear subject separation
  4. Classify the thumbnail job:
    • talking-head / opinion / tutorial
    • product / software / demo
    • cinematic / documentary / art
    • entertainment / challenge / reaction
    • education / business / concept
  5. Propose concise title text options.
  6. Offer 3-6 clearly different style directions.
    • Each option should name the composition, not only a channel/style label.
    • Exclude weak options.
    • Wait for user choice when the direction is not obvious.
  7. Generate thumbnail options using selected real frames and the selected style.
  8. Check mobile readability, subject fidelity, product/UI truthfulness, and visual contrast.
  9. Present results and offer targeted adjustments.

Prompt Shape

[Platform] thumbnail, [aspect ratio / dimensions], [style name].

COMPOSITION: [subject position, scale, background, graphic balance]
REAL REFERENCE: [describe selected frame zone by zone]
TEXT: "[short title]" in [high-contrast style]
STYLE: [one-line style recipe from references]
CONSTRAINTS: readable at mobile size, no fake UI, no watermark, no play button.

References

  • references/thumbnail-styles.md: style library from real thumbnail analysis.
  • references/thumbnail-style-recipes.md: quick style recipes and prompt patterns.
  • references/field-testing-notes.md: known failure modes.

Signals

GitHub stars
2k
Forks
265
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
video-thumbnail-generator
Source
github.com/0xsline/openchatcut