clean-infographics ✔ security-verified · 3 models

---
name: clean-infographics
purpose: Generate clean, legible AI infographics that never come out with garbled text — title-only rendering, verified by OCR before delivery.
category: media
tags: [infographics, image-generation, design, verification, dataviz]
license: CC0
---

# Clean Infographics — the title-only formula

AI image models garble long text. This skill produces professional infographics that survive
that limitation, verified before they ever reach a human.

## The core rule

**The image carries ONE piece of text: the title. Everything else is wordless visual metaphor.**
Detail lives in the caption/post next to the image, never inside the pixels.

## Procedure

1. **Write the prompt with this locked structure:**
   - Style line: "{{BRAND_STYLE}} infographic poster, {{BRAND_COLORS}} palette, lots of negative space"
     (e.g. "warm hand-drawn artisan poster, cream paper, gold and brown ink").
   - Exactly TWO focal visual metaphors, described concretely ("a loaf of bread with a trail of
     crumbs", "two hands exchanging a glowing crumb"). More than two = clutter.
   - The text clause, verbatim: `The ONLY text is the title '<TITLE>'. No other words, labels,
     numbers, foreign characters, or gibberish. English only.`
2. **Generate** with your image model ({{IMAGE_MODEL}} via {{IMAGE_API}}), 1024x1024 or larger.
3. **Verify with your own eyes/OCR before delivering** — the non-negotiable step:
   - The title is spelled exactly right.
   - There is NO other text, watermark fragments, or pseudo-letters anywhere.
   - Both metaphors are present and recognizable.
   If any check fails, regenerate with this appended: "Make the composition simple and bold.
   Render the title large, crisp and perfectly legible with correct spelling; keep everything
   else wordless." Two failures → change the metaphors, not the rule.
4. **If data must be shown** (charts, scores, steps): do NOT ask the image model to draw it.
   Render it programmatically (HTML/SVG/PIL) using dataviz basics — one hue for magnitude,
   direct labels, no dual axes — and composite or attach it separately.
5. **Deliver** the image plus a 1-3 line caption carrying the details the image doesn't.

## Configure once ({{PLACEHOLDERS}})

- {{IMAGE_MODEL}} / {{IMAGE_API}} — your image endpoint (any GPT-image-class model works).
- {{BRAND_STYLE}} / {{BRAND_COLORS}} — your look; keep it consistent across every infographic.
- {{FONT_SIZE_MIN}} — accessibility floor for any programmatic text overlays.

## Why it works

Text failure is the #1 reason AI infographics look amateur. Removing all text but the title
eliminates the failure surface; OCR verification catches the rest. Metaphors do the explaining,
the caption does the detail. Eval: generate 3 posters on different topics — all 3 must pass
step 3 on the first or second try.