| name | create-infographics |
| description | Generate publication-quality infographics from any markdown content — talk summaries, research papers, product plans, blog posts, dashboards.
Default visual style: modern enterprise AI / SaaS keynote (white background, blue + orange accents, rounded cards, clean sans-serif, generous whitespace). Optional overrides: chalkboard, aged-academia, technical-schematic, pop-laboratory, morandi-journal.
Standard outputs: 16:9 landscape infographic per topic/section, 9:16 portrait poster for whole-content overviews, 3:4 portrait for dense-modules infopedia.
Trigger when the user asks to "create infographic", "generate infographic", "make a poster", "visualize this content", or provides a markdown file and asks for a visual summary.
|
| metadata | {"origin":"derived from llm-wiki-skill project","generation_engine":"baoyu-image-gen (bundled in deps/)","default_provider":"google · gemini-3-pro-image-preview","default_style":"executive-saas"} |
Create Infographics
Turn any markdown content into a publication-quality infographic image. Default style is executive-saas — modern AI / SaaS keynote aesthetic with white background, enterprise blue + warm orange accents, rounded cards with subtle shadows, generous whitespace, and clean sans-serif typography. Override the style only when the user explicitly names a different aesthetic.
When To Use This Skill
- User provides markdown content (talk summary, paper, product plan, blog post, etc.) and asks for an infographic
- User says: "create infographic", "generate infographic", "visualize this", "make a poster", "make this a long infographic"
- User asks for a presentation-ready visual from existing notes
Do NOT trigger when:
- User just wants the content summarized in text (no visual)
- User wants a chart or technical diagram (use a charting tool instead)
- User wants a UI mockup of a real product (use design tools)
- User wants a photo, illustration, or creative artwork (use generic image gen directly)
Required Tools & Setup
The skill bundles two dependencies under deps/:
baoyu-image-gen — the image generation engine (Google Gemini default; OpenAI, OpenRouter, DashScope, Replicate also supported)baoyu-infographic — reference content (21 layouts × 20 visual styles) used to ground layout decisions
Prerequisites
bun (preferred) or npx runtime for baoyu-image-gen- API key for at least one image-gen provider (defaults to
GOOGLE_API_KEY)
- Provider preferences set in
~/.baoyu-skills/baoyu-image-gen/EXTEND.md (the bundled skill prompts for first-run setup)
See INSTALL.md for one-command setup.
The Workflow
Step 1 — Read the source content
Read the markdown file the user pointed at. Identify:
- Content shape: does it have a transformation arc (current → future)? Multiple parallel themes? A single dense argument? A maturity progression?
- Section count: how many distinct claims, themes, or stages does the content have?
- Key data points: numbers, quotes, tables that should be visually emphasized
Step 2 — ALWAYS ASK: which density level?
This step is MANDATORY for whole-content posters (talk / paper / video / book summaries). Use AskUserQuestion to present the two density tiers, unless the user already specified one in their prompt (e.g., "make a dense 3:4 poster" or "use the normal density").
Density tiers:
| Tier | Aspect | Per-module content | Use when |
|---|
| High density (infopedia) | 3:4 portrait | 2-4 sub-elements per module (mixed interior patterns: 2-col comparisons, 3-step flows, 2x2 grids, hero-stats) | Whole-talk recap with lots of substance; reference posters; users who want to cover all content |
| Normal density (breathing-room) | 9:16 portrait | 1 element per section (single stat, single bullet list, single quote) | Light or narrative content; executive keynote scan-in-5-seconds; users who want a clean cover image |
Always present this choice with AskUserQuestion, header label "Density", with High recommended for substantive source content and Normal recommended for thin or narrative content. Skip the question only if the user's request already named a density or aspect ratio.
Step 3 — Match content shape to layout
Use references/layouts.md to pick the right layout.
Quick decision tree:
| Content shape | Layout | Aspect |
|---|
| Current → future transformation, 4 stages | 4-stage maturity arc | 16:9 landscape |
| 4-7 parallel themes / cards | bento-grid or n-card horizontal | 16:9 landscape |
| Side-by-side comparison (A vs B, before vs after) | binary-comparison | 16:9 landscape |
| Single deep dive on one framework | dashboard or hub-spoke | 16:9 landscape |
| Whole-content summary, HIGH density chosen | dense-modules infopedia (multi-element sub-content per module) | 3:4 portrait |
| Whole-content summary, NORMAL density chosen | dense-modules (single-element sections) | 9:16 portrait |
| Per-vendor / per-section detail | one-card per item | 16:9 landscape |
Step 4 — Pick the style
Default: executive-saas — apply unless the user explicitly requests another style.
Override only when:
- User asks for
chalkboard (workshop / classroom feel)
- User asks for
aged-academia (vintage scientific journal)
- User asks for
pop-laboratory (technical precision + pop accents)
- User asks for
morandi-journal (warm hand-drawn doodle)
- User asks for
technical-schematic (engineering blueprint)
- User asks for any of the other 20 styles in
deps/baoyu-infographic/references/styles/
Step 5 — Build the prompt
Use the template in references/prompt-template.md. The template has 4 standard zones:
- Top header strip (master title + subtitle + tiny attribution)
- Hero / thesis band (full-width statement)
- Main content (the layout-specific zone — cards, stages, comparisons, etc.)
- Bottom takeaway banner (dark slate, white text)
ALWAYS include the anti-duplication safeguards from references/anti-duplication.md at the top of the prompt. Gemini is known to duplicate section numbers in dense vertical layouts unless explicitly told not to.
Step 6 — Generate the image
Call the bundled baoyu-image-gen via batch file:
${BUN_X} {skill-dir}/deps/baoyu-image-gen/scripts/main.ts --batchfile <path-to-batch.json> --json
Standard batch file shape:
{
"jobs": 3,
"tasks": [
{
"id": "overall-poster",
"promptFiles": ["prompts/overall.md"],
"image": "overall.png",
"provider": "google",
"model": "gemini-3-pro-image-preview",
"ar": "9:16",
"quality": "2k"
}
]
}
Step 7 — Verify and iterate
Common Gemini quality issues to check for:
- Duplicated section badges (e.g., two "04" sections) — fix by tightening anti-duplication language
- Illegible small text at 9:16 with dense layouts — fix by reducing section count
- Layout drift (sections rearranged or missing) — fix by adding explicit ordering ("EXACTLY 6 numbered sections, in this order: 01, 02, 03, 04, 05, 06")
- Garbled equations or special characters — fix by spelling out (e.g., "capability multiplied by taste multiplied by agency" instead of
×)
If the image is close but a single zone is wrong, regenerate only that zone (modify the prompt) — don't reroll the whole image.
Output Locations
Save outputs alongside the source MD by default:
docs/
└── source-content.md
├── prompts/
│ ├── overall.md
│ ├── section-01.md
│ └── section-02.md
├── batch.json
├── overall.png
├── section-01.png
└── section-02.png
Use a prompts/ subdirectory so iteration is cheap — the user can edit a prompt and re-run just that task.
Examples
See examples/ for worked examples:
Always ask which density tier (HIGH or NORMAL) the user wants before generating a whole-content poster — see Step 2 of the workflow above.
References
For the underlying tools:
