Self Media Tech Thumbnail

Turn a topic or block of copy into a ready-to-generate thumbnail prompt, then generate the image when the user asks for a cover. Default output is a 16:9 landscape thumbnail optimized for Chinese self-media feeds.

Workflow

1. Open references/style-system.md when the request says "类似这种风格", "照这个做", or needs a closer match to the bundled example.
2. Default to 16:9 unless the user names another format.
3. Compress the user's copy into a three-level text hierarchy:
   • kicker: 4-10 Chinese characters, usually authority, urgency, or surprise
   • main headline: 8-18 Chinese characters, optionally keep one English product name for recognition
   • support line: 8-18 Chinese characters, usually contrast, result, conflict, or payoff
4. Choose the focal visual:
   • Use a person when the topic centers on a founder, engineer, or public figure.
   • Use product UI, code, or interface cards when the story centers on a tool or feature.
   • Use a versus layout only when the story is explicitly comparative.
5. Build the image prompt with the style rules below.
6. If the user asks for a cover image, thumbnail, or封面图, call image_gen directly. If the user asks only for文案 or提示词, return the text pack and the generation prompt.

Style Rules

• Make the composition bold and asymmetrical. Default to a heavy title block on the left and the main subject on the right.
• Use a dark graphite, navy, or black background with electric blue plus warm orange or red accent lighting.
• Make the main Chinese headline oversized, condensed, and extremely readable on mobile.
• Favor white for the core Chinese claim, yellow for the key English product word, and cyan or blue for secondary UI text.
• Add 2-4 supporting modules such as a checklist card, code panel, feature strip, product UI card, or comparison badge.
• Use depth, drop shadow, glow, rim light, and diagonal motion cues to create a premium breaking-news feel.
• Keep the image dense but controlled. Every supporting element must reinforce the headline.
• Prefer photorealistic人物, crisp UI, sharp typography, and clean edge lighting.

Avoid

• Do not recreate the bundled reference image verbatim.
• Do not rely on soft pastel palettes, tiny text, or flat minimal layouts.
• Do not let visual effects bury the main headline.
• Do not exceed four text blocks.
• Do not use exact logos or celebrity likenesses unless the user explicitly asks for them.

Prompt Template

Use a prompt in this shape and fill in the content:

Create a bold Chinese self-media AI/tech video thumbnail, 16:9, optimized for mobile feed readability. Dark high-contrast cinematic background with electric blue and warm orange/red lighting, premium editorial tech aesthetic, oversized condensed Chinese headline text with strong depth and shadow. Main focal subject: <subject>. Left-heavy text composition, right-side hero visual, supporting UI panels and code/product elements, subtle lightning or motion streaks, dramatic contrast, sharp edges, glossy highlights.

Headline text:
Top kicker: "<kicker>"
Main headline: "<main headline>"
Support line: "<support line>"
Optional corner line: "<optional short CTA or trend line>"

Visual cues: <2-4 supporting elements>.
Mood: breaking-news, high authority, high click-through, polished, not cluttered.
No watermark, no unrelated text, no blurry typography.

Copy Compression

When the user gives long copy, reduce it to:

• the claim
• the object
• the tension

Examples:

• claim: 官方下场
• object: Claude 提示词
• tension: 为什么很多人还是写不好

If the user provides only a topic, invent concise cover copy that sounds native to Chinese self-media, then build the prompt from that.

Output Modes

• If the user asks for 封面, 缩略图, 出图, or 来张图, generate the image directly.
• If the user asks for 标题, 封面文案, or 提示词, return:
  • one primary headline pack
  • one ready-to-use image prompt
• If the user asks for several options, vary the hook and support line while keeping the same visual system.

Bundled Resources

• Style notes: references/style-system.md
• Visual anchor: assets/reference-grok-build.png

Open the bundled style notes when you need a refresher on composition, color, or module choices. Use the reference image only as a style anchor, not as a frame to copy element-for-element.