Comics Maker
guyaga/10d10s-bonus-comics-maker
Crie quadrinhos print-ready com arte IA consistente em minutos
Ver no GitHubSkill que transforma um conceito e uma foto sua em uma história em quadrinhos completa, pronta para impressão. Planeja, gera arte com Gemini 3 Pro (mantendo consistência facial em cada painel), adiciona textos e vincula tudo em PDF. Ideal para criar narrativas visuais coerentes onde o personagem principal permanece reconhecível do início ao fim.
Instalação
npx convez-ai guyaga/10d10s-bonus-comics-maker1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Gerar histórias em quadrinhos personalizadas com seu rosto como protagonista
- Criar portfólios visuais ou narrativas de marketing em formato de HQ
- Produzir material de treinamento ou storytelling empresarial em quadrinhos
Veredito

Ferramenta bem arquitetada que resolve o problema real de consistência facial em gerações repetidas de IA. Pipeline claro (planejar → gerar → letreirar → vincular) e integração nativa com Claude Code a tornam acessível. Útil para criadores de conteúdo e aplicadores em narrativa visual.
README
A tradução para o português deste README ainda está na fila.
📖 Bonus: Comics Maker
10 Days 10 Skills — course bonus Turn a one-line concept (and your own face) into a full, print-ready comic book. Plan → generate art → letter → bind PDF. The hero stays recognizable on every page.
This is the skill behind "The Debugger" — a 12-page cyberpunk issue where a real person becomes a code-bending hero in Tel Aviv 2049. You describe the comic in one config file; three small Node scripts do the rest. A reference photo is fed into every generation call, so the face, hair, build and signature outfit stay consistent across all twelve pages.
It builds on the course's image day:
| Step | Tool | Course day |
|---|---|---|
| Plan the issue | COMIC_PLAN.md (character sheet + page-by-page script) | — |
| Generate the art | Gemini 3 Pro Image (fast) or GPT Image via the Codex CLI (sharper, free on a ChatGPT login) — photo on every panel | Day 1: Image Generation |
| Letter the pages | sharp + hand-built SVG (classic white caption boxes, speech bubbles, SFX) | — |
| Bind the book | pdfkit → US-comic-sized PDF | — |
Two art backends (backend: "gemini" | "gpt2")
gemini(default) — Gemini 3 Pro Image. Fast (~seconds/page). NeedsGEMINI_API_KEY.gpt2— drives the Codex CLI's built-inimage_gentool (codex exec -m gpt-5.5). Noticeably sharper line work and crisper in-art text (titles, SFX, signage), ~1–2 min/page, and free on a ChatGPT-accountcodexlogin (noOPENAI_API_KEYbilled). The reference photo is passed with-iso the hero's face carries across every page.
🔁 The pipeline
plan (COMIC_PLAN.md) → comic.config.mjs → generate.mjs → assemble.mjs → make-pdf.mjs → <title>.pdf
- Plan — interview the user, write
COMIC_PLAN.md(concept, hero, art style, page breakdown). - Config — author
comic.config.mjs: astylePrefix+charDescreused on every prompt, then one entry per page (prompt + text overlays). - Generate —
scripts/generate.mjsmakes the character sheet + every page, feeding the reference photo to each call. Re-run a single page with--only=page04. - Letter —
scripts/assemble.mjscomposites caption boxes / speech bubbles / SFX onto the art (pages where the model already drew the text just get a page number). - Bind —
scripts/make-pdf.mjslays every page onto a comic-sized PDF.
🚀 Quick start
git clone https://github.com/guyaga/10d10s-bonus-comics-maker
cd 10d10s-bonus-comics-maker
cp .env.example .env # add your GEMINI_API_KEY
npm install
# set up a comic project
mkdir -p "My Comic/Assets"
cp reference/comic.config.mjs "My Comic/comic.config.mjs" # then edit it
# drop your hero photo at "My Comic/Assets/hero-reference.jpg"
export GEMINI_API_KEY=your-key # PowerShell: $env:GEMINI_API_KEY="your-key"
# run the three stages
node scripts/generate.mjs "My Comic"
node scripts/assemble.mjs "My Comic"
node scripts/make-pdf.mjs "My Comic"
# -> "My Comic/The_Debugger_Issue_1.pdf"
Installed as a Claude Code skill (in
~/.claude/skills/comics-maker/), just say "make me a comic about …" and Claude runs the whole pipeline for you.
🔑 Keys
| Key / login | For | Required? |
|---|---|---|
GEMINI_API_KEY | the gemini backend (Gemini 3 Pro Image) | for gemini |
codex CLI logged in (ChatGPT account) | the gpt2 backend (free built-in image_gen) | for gpt2 |
OPENAI_API_KEY | only gpt2 with mode:"api" (paid gpt-image-2) | optional |
🧱 Config in one screen
comic.config.mjs exports the whole comic. Define the two consistency constants once,
reuse them everywhere:
const stylePrefix = "Modern graphic novel comic book art style. ...";
const charDesc = 'The main character looks exactly like the reference photo: ...';
export default {
title: "The Debugger - Issue 1",
refPhotos: ["Assets/hero-reference.jpg"], // fed into every page
defaultAspect: "3:4",
pageSize: { widthIn: 6.625, heightIn: 10.25 },
characterSheet: { file: "sheet.png", aspect: "16:9", prompt: `${stylePrefix} ...` },
pages: [
{
id: "page01", folder: "Page_01", file: "page01.png",
useRef: false, // no hero on this page
prompt: `${stylePrefix} FULL PAGE SPLASH ...`,
overlays: [{ type: "caption", text: '"Tel Aviv. 2049."', x: 0.03, y: 0.82, width: 820 }],
},
// ...one entry per page
],
};
Overlay types: caption (purple narrator box), speech (white bubble w/ tail), sfx
(red glitch text), title, subtitle. Positions are fractions of the page (0–1).
See reference/comic.config.mjs for the complete 12-page "Debugger" example and
reference/COMIC_PLAN.md for the planning doc that precedes it.
💡 Hard-won lessons (baked into the skill)
- Feed the reference photo to every hero panel. This is the single biggest lever on face consistency — far more than describing the face in words.
- Reuse the style prefix verbatim on every prompt, or the art style drifts page to page.
- Let the model draw the big lettering (SFX, title cards, signage) inside the art, and add only clean narrator captions / speech bubbles afterward. Mixing both reads best.
- Position overlays with fractions, not pixels, so they survive any output resolution.
- Splash pages (1, 7, 12) carry the issue — spend the most prompt care there.
- A page came out wrong? Tweak its prompt and re-run just that page with
--only=— no need to regenerate the whole issue.
Course bonus for 10 Days, 10 Skills. MIT licensed.