<< All versions

Skill v1.0.0

currentLLM-judged scan90/100
0xsline/openchatcut/video-thumbnail-generator
──Details
PublishedOctober 20, 2026 at 09:24 PM
Content Hashsha256:77bcf1f26ec6f822...
Git SHA23ff26c0c929
──Files
Files (1 file, 2.8 KB)
SKILL.md2.8 KBactive
SKILL.md · 70 lines · 2.8 KB

version: "1.0.0" name: video-thumbnail-generator description: 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.


Video Thumbnail Generator

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.
  1. 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.
  1. 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.
  1. Read project state, transcript, and available media context.
  2. Inspect candidate frames:
  • expressive face
  • product or UI clearly visible
  • dramatic action moment
  • clean composition
  • strong lighting or clear subject separation
  1. Classify the thumbnail job:
  • talking-head / opinion / tutorial
  • product / software / demo
  • cinematic / documentary / art
  • entertainment / challenge / reaction
  • education / business / concept
  1. Propose concise title text options.
  2. 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.
  1. Generate thumbnail options using selected real frames and the selected style.
  2. Check mobile readability, subject fidelity, product/UI truthfulness, and visual contrast.
  3. Present results and offer targeted adjustments.

Prompt Shape

text
[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.
All versions