Baoyu Visual Skills

Baoyu Visual Skills — Infographics, Comics & Article Illustrations

Three visual content generators from the baoyu suite (by 宝玉/JimLiu), all sharing the same image_generate-based backend, prompt-only workflow, and file structure conventions.

Section 1: Infographics (baoyu-infographic)

21 layouts × 21 styles for creating information graphics, data visualizations, and educational diagrams.

When to Use

User asks for infographic, visual summary, information graphic, "信息图", "可视化", or "高密度信息大图".

Layouts (21)

Styles (21)

craft-handmade (default), claymation, kawaii, storybook-watercolor, chalkboard, cyberpunk-neon, bold-graphic, aged-academia, corporate-memphis, technical-schematic, origami, pixel-art, ui-wireframe, subway-map, ikea-manual, knolling, lego-brick, pop-laboratory, morandi-journal, retro-pop-grid, hand-drawn-edu

Workflow

1. Analyze content → save source + analysis files
2. Recommend 3-5 layout × style combinations based on content structure
3. Confirm with user (combination, aspect ratio, language)
4. Generate prompt using layout + style definitions → save to prompts/infographic.md
5. Generate image via image_generate → download URL to output directory
6. Report summary

Output structure: infographic/{topic-slug}/ with source, analysis.md, structured-content.md, prompts/, and final PNG.

Keyword Shortcuts

• high-density-info / 高密度信息大图 → auto-select dense-modules + morandi-journal/pop-laboratory/retro-pop-grid
• infographic / 信息图 → auto-select bento-grid + craft-handmade

Section 2: Knowledge Comics (baoyu-comic)

Create educational/educational comics with flexible art style × tone combinations.

When to Use

User asks for knowledge comic, 知识漫画, educational comic, biography comic, tutorial comic, or Logicomix-style.

Art Styles (6)

Tones (7)

neutral (default), warm, dramatic, romantic, energetic, vintage, action

Presets (5)

Layouts

standard (default), cinematic, dense, splash, mixed, webtoon, four-panel

Workflow

1. Analyze content → save source + analysis
2. Confirm style, tone, layout, aspect ratio with user
3. Generate storyboard + character definitions
4. Generate prompts per page (with character descriptions embedded)
5. Generate images via image_generate → download each to local PNG
6. Character sheet optional (human review artifact)

Output structure: comic/{topic-slug}/ with source, analysis.md, storyboard.md, characters/, prompts/, and PNG pages.

Character Consistency

• Character descriptions in characters/characters.md text format
• Embedded inline in every page's image_generate prompt
• PNG character sheet optional (human-facing, not input to image model)

Section 3: Article Illustrations (baoyu-article-illustrator)

Analyze articles, identify illustration positions, generate images with Type × Style × Palette consistency.

When to Use

User asks to illustrate an article, add images, generate illustrations for content, "为文章配图", or "illustrate article".

Types (6)

Styles

Core: notion-style, warm-illustration, minimal-flat, tech-blueprint, watercolor, elegant-line

Full style gallery: references/styles.md

Palettes (optional)

macaron, warm, neon — override style's default colors. Or define custom palette.

Workflow

1. Detect reference images (analyze via vision_analyze for style traits)
2. Analyze content (type, purpose, core arguments, positions for illustrations)
3. Confirm settings (preset/type, density, style, palette, language)
4. Generate outline with per-illustration entries
5. Generate prompts per illustration
6. Generate images via image_generate → download each
7. Insert ![](relative/path) into article after corresponding paragraph

Output structure: {output-dir}/ with source, analysis.md, outline.md, prompts/, and PNGs.

Shared Conventions

File Structure

{type}/{topic-slug}/
├── source-{slug}.{ext}
├── analysis.md
├── prompts/
│   └── NN-{type}-{slug}.md
└── NN-{type}-{slug}.png

Aspect Ratio Mapping

Core Principles (All Baoyu Skills)

1. Data integrity — never summarize or paraphrase source statistics
2. Strip secrets — scan for API keys, tokens, or credentials before writing any output
3. Prompt files are mandatory — save prompt to disk before every image_generate call
4. Always download the URL returned by image_generate to a local file (the tool returns a URL, not a file)
5. Use absolute paths for curl — never trust persistent-shell CWD across batches
6. Backup before regenerating — existing files get -backup-YYYYMMDD-HHMMSS suffix

clarify Timeout Handling

When clarify times out (no user response):

• Use best judgment for THAT question only
• Continue asking remaining questions if more are needed
• Surface the default visibly so user can correct later

Pitfalls

1. image_generate does not accept images — it's prompt + aspect ratio only. Reference images must be described textually via vision_analyze.
2. Always download URLs — image_generate returns a URL, not a local path. Use curl -fsSL "<url>" -o /abs/path/to/file.png
3. Absolute output paths — never relative paths for curl -o; CWD can drift between batches
4. No backend selection — image_generate uses whatever model the user configured; don't write model names into prompts
5. One concept per section — each illustration/comic page/infographic section should convey one clear concept
6. Comic Step 2 confirmation required — don't skip style+tone confirmation
7. Metaphors vs concepts — illustrate the underlying concept, not the literal metaphor
8. Character consistency is text-driven — embed character descriptions in every page prompt