YouTube Thumbnail

---
name: 'youtube-thumbnail'
description: 'High-click thumbnails: big expressive subject, bold contrast and a 2-4 word hook that reads at small sizes.'
---

# YouTube Thumbnail

Makes a 16:9 thumbnail built to read at phone size, with one subject, one emotion and one idea.

## Use it for

- YouTube videos, podcasts and course lessons
- Testing several thumbnail ideas for one video

## Inputs

- Video topic
- Optional: 2-4 word hook text, a photo of the host (their own face only)

## Defaults

- Call `generate_image`. If the user attaches their own photo, pass it in `reference_images` and keep their likeness.
- Aspect ratio 16:9.
- Subject on one side, text on the other.

## Prompt recipe

`YouTube thumbnail, [subject with exaggerated emotion] on the [left/right], [key object or visual metaphor], bold saturated background with strong contrast, rim light separating subject from background, text "[HOOK]" in huge bold letters with a thick outline on the other side, clean and readable at small size`

## Example prompt

```text
YouTube thumbnail, a shocked young man with wide eyes on the left pointing at a giant cracked smartphone, bold yellow-to-orange background with strong contrast, rim light separating him from the background, text "DON'T BUY" in huge bold white letters with a thick black outline on the right, clean and readable at small size
```

## Variations to offer

- Three different hooks for testing
- A no-face version with an object as the hero
- A darker, mysterious version

## Avoid

- More than 4 words of text
- Using someone else's face
- Small details that disappear when shrunk

## Working rules

- Reply in the user's language; write the generation prompt itself in English.
- Ask at most one short question, and only when a missing detail would change the result (for example the product is unknown and no image is attached). Otherwise use the defaults in this skill and go.
- Keep one generation call per requested output. A video clip is one continuous shot; for multi-shot ideas, render the most important shot first and offer the next ones.
- Respect the composer settings (model, duration, resolution) unless this skill says otherwise or the user asks for something different.
- Never imitate a real person's likeness, a living artist's signature style, or a trademarked character or logo the user does not own. Swap them for an original description.
- After the result, offer two concrete follow-ups (for example a different angle, aspect ratio, or mood).