baoyu-comic — Turn Any Knowledge Into a Manga

baoyu-comic is a Claude Code skill that converts any topic — a concept, biography, or tutorial — into a multi-page knowledge manga with consistent characters, a storyboard, and exported pages. It ships with six art styles, seven tones, seven layouts, and five named presets, making it a full visual-content pipeline inside an AI agent. (The original baoyu-skills version can merge pages into a PDF; the Hermes port output format varies — see repo for current behavior.)

Who Built This

The skill was created by JimLiu, known online as 宝玉 (Baoyu), Twitter @dotey, based in Chicago (site: baoyu.io). His baoyu-skills repo has 19,100+ GitHub stars and 2,200+ forks. baoyu-skills is a widely-used Claude Code skill collection (github.com/JimLiu/baoyu-skills). The skills are written in TypeScript and run via Bun.

The skill was ported into Nous Research’s Hermes Agent (open-source) as a built-in bundled skill. Hermes runs on 200+ models via OpenRouter/Anthropic/OpenAI and compatible local backends; baoyu-comic is included as a built-in skill (see hermes-agent.nousresearch.com for current skill catalog and version details).

Installation

# Via baoyu-skills (original — standalone install)
npx skills add https://github.com/jimliu/baoyu-skills --skill baoyu-comic

# Via Hermes Agent (bundled built-in — no install needed)
# Trigger keywords: "知识漫画", "教育漫画", "biography comic", "tutorial comic", "Logicomix-style"

Install via Hermes Agent directly (bundled built-in skill); for standalone use, install from the baoyu-skills collection via npx.

Requires: an image generation API key — OpenAI, Azure OpenAI, Google, OpenRouter, DashScope, or Replicate.

The Six Art Styles

The Seven Tones & Seven Layouts

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

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

Five Named Presets

Presets combine style + tone + layout into a single opinionated package:

CLI Flags

--art          Art style (see table above)
--tone         Emotional tone
--layout       Panel layout
--aspect       Image ratio: 3:4 portrait (default), 4:3, 16:9
--lang         Output language
--ref          Reference image for character consistency
--batch-size   Pages per batch render
--storyboard-only   Stop after storyboard generation
--prompts-only      Generate image prompts but skip rendering
--images-only       Skip storyboard, render from existing prompts
--regenerate N      Re-render page N

The Full Pipeline

1. Load user preferences from EXTEND.md (custom style rules)
2. Analyze content + confirm style with user
3. Generate storyboard + character definitions
4. Write image prompts as .md files (one per panel — reproducibility record)
5. Generate a character sheet PNG (ensures consistent appearance across pages)
6. Batch-render pages via the image generation backend
7. Merge all pages into a final PDF

Output folder: comic/{topic-slug}/ containing source markdown, content analysis, storyboard, character definitions, per-panel image prompt files, PNGs, and the final PDF (original version). The Hermes port emphasizes PNG downloads and a completion report; PDF merging behavior may differ.

Hermes Port vs. Original

How LearnAI Team Could Use This

• AI Education outreach — convert a complex topic (e.g., “what is a transformer?”) into a 6-page manga that non-technical audiences actually want to read and share
• Course material previews — create a manga teaser for a new module to boost enrollment interest and social reach
• Research communication — turn a paper’s key findings into a Logicomix-style narrative for conference posters or X/LinkedIn posts
• Student assignments — ask students to generate a knowledge manga as a synthesis project; the storyboard phase forces structured conceptual thinking
• Showcasing Claude Code capabilities — live-demo the skill to show how far agent workflows have come beyond text output

Real-World Use Cases

Important Things to Know

• Image generation costs money. Each page is a separate API call. Use --storyboard-only or --prompts-only to preview before committing to a full render.
• Character consistency is the hard problem. The skill addresses this with a dedicated character sheet PNG and text descriptions, but photorealistic styles show more drift across pages than manga or minimalist.
• The storyboard is the real work. A good storyboard means a good comic. The text analysis + storyboard phase is where the AI does the most valuable structuring — worth reviewing before auto-rendering.
• --regenerate N is your friend. If one panel doesn’t land, regenerate just that page rather than re-running the whole batch.
• EXTEND.md is your customization hook. Drop persistent style rules, character defaults, or brand preferences into EXTEND.md and the skill picks them up automatically on each run.