# Vertical AI Agent Knowledge Card Prompt Recipe

> Status: reference prompt. Use this when generating 4:5 AI Agent course knowledge cards with Codex / ChatGPT Pro image generation.

## What This Produces

- 4:5 vertical educational infographic cards for course chapters
- Warm Raymond brand palette
- Dense but readable Traditional Chinese information layout
- Bottom-right minimal watermark: `雷蒙三十｜lifehacker.tw`

## Base Prompt

```text
Create a PORTRAIT infographic card with EXACT aspect ratio 4:5: width 1122, height 1402. The image must be taller than wide, but only slightly: 4:5 portrait, NOT landscape, NOT square, NOT 9:16, NOT 2:3, NOT tall skinny. The canvas should feel broad and spacious like an Instagram portrait carousel card.

Style: polished AI-generated Taiwanese educational knowledge-card infographic, same style as the successful original cards: large Traditional Chinese headline, top-left numbered badge, clean rounded panels, arrows, small line icons, thin dividers, bottom keyword chips, high information density but readable.

Brand palette only: warm cream #F7F5F0 background, teal #21A4B1 and #007481, action orange #E77E47, yellow #F6B85A, ink #111111/#3A3A3A.

IMPORTANT fixed watermark: bottom-right must contain only minimal plain text exactly「雷蒙三十｜lifehacker.tw」. No avatar, no logo icon, no rounded rectangle, no sticker, no border, no extra words.

Card NN / NN. Topic:「{topic}」
Main title exactly:「{main_title}」
Subtitle:「{subtitle}」
Core visual: {central_visual}
Required panel text:
1「{panel_1}」
2「{panel_2}」
3「{panel_3}」
4「{panel_4}」
Bottom keyword chips:「{chip_1}」「{chip_2}」「{chip_3}」「{chip_4}」

Avoid: landscape layout, tall skinny layout, cramped text, misspelled Chinese, random English, purple palette, avatar in watermark, boxed watermark, 3D mascot, photorealistic people.
```

## Notes

- Put the ratio instruction at the very top. Without this, the model may generate a tall skinny 9:16-like card.
- Keep the watermark instruction explicit and negative: no avatar, no boxed watermark, no extra course name.
- For text-heavy cards, prefer fewer required text strings and let the model use visual hierarchy. Too many exact labels increase typo risk.
- Use selected output examples:
  - `ai-agent-vertical-card-handbook-4x5.png`
  - `ai-agent-vertical-card-memory-4x5.png`