# Visual Creation

> Knowledge graphic generation methodology — 5 style families, structured prompt template, validation checklist. For turning any text/speech/review into a publishable visual card via AI direct generation (no PPT, no HTML conversion needed).

- Skill: `huangrichao2020/visual-creation` (Agent Skill, multi-file: 38 files)
- Install (CLI): `npx skillmds@latest add huangrichao2020/visual-creation`
- Raw SKILL.md: https://api.skillmd.com/api/skills/huangrichao2020/visual-creation/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Web & Frontend
- Author: huangrichao2020 (https://skillmd.com/u/huangrichao2020)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/huangrichao2020/visual-creation

---


# Knowledge Graphic Generation

Turn any content into publishable visual cards via AI direct image generation. Not PPT, not HTML — straight from prompt to PNG. The dominant format for knowledge propagation in 2026.

## Core Positioning

```text
Other knowhub domains: answer "what is this domain's knowledge/judgment"
visual-creation: answer "how to turn this content into a propagatable graphic"
```

## When to load

- User says "knowledge graphic" / "info graphic" / "knowledge comic" / "PPT is dead" / "make me a Xiaohongshu card"
- User shares a text/speech/outline and asks for a visual version
- User wants to iterate a poster/cover/infographic style

## 5 Style Families

See `references/style-catalog.md` for full prompt recipes.

| Style | One-liner | Best for |
|---|---|---|
| Corporate Tech | Blue-white-gray + line icons + rounded cards | Industry reports, tech retrospectives |
| Museum/Catalog | Light gray-blue + soft gradient + modular | Annual reports, knowledge matrices |
| Hand-drawn Edu | Macaron + doodle + journal layout | Xiaohongshu, video accounts, education |
| Real Life | Sticky note / paper texture + natural handwriting | Emotional content, personal IP |
| City Illustration | Ink wash + landmark silhouettes | Cultural products, postcards |

## Universal Prompt Template (5 blocks)

```
[1 SCENE]    What type of graphic (infographic / poster / map / storyboard)
[2 SUBJECT]   Core content (one sentence)
[3 STRUCTURE] How many blocks, what in each (numbered 1./2./3.)
[4 STYLE]    Style keywords + palette + font + composition + decorations
[5 CONSTRAINT] No typos, language, resolution, banned elements
```

## 4 Hard-won Tips

1. **Structured prompts** — Block-based prompts are 5x more stable than long paragraphs
2. **Reverse-engineer style** — Describe a liked image's style → feed description back as new prompt
3. **Number complex info** — 1./2./3. lists keep AI from missing modules
4. **Always human-verify** — Occasional typos / missing elements are normal; regenerate or local retouch

## Quality Bar

```text
□ Text renders accurately (no garbled CJK, no missing/extra chars)
□ All listed modules present
□ Style consistent across the graphic
□ Main title visible in 2 seconds
□ Information density balanced (one screen = one core takeaway)
□ Publish-ready (Xiaohongshu / WeChat / video account)
```

## Available Tools

- **AI direct generation**: matrix MCP (GPT Image2-class model, accurate Chinese rendering) — **默认 PPT 路线**
- **python-pptx 嵌入图片** → 输出真实 .pptx（PowerPoint / Keynote / WPS 可打开）— PPT 任务默认路线
- ~~HTML precision version~~ — **已废止**（2026-07-10 PPT 任务用户偏好 v2 固化，HTML 阅读器路线默认不走）
- ~~Hybrid (AI 背景 + HTML 文字层)~~ — **已废止**（同上）