LlamaGen Logo

Claude Code tutorial · 獨立創作教學

Claude Comic Book Skill:從劇本到漫畫分鏡

建立 SKILL.md,定義角色設定、頁面、分格和連續性規則。真正的漫畫畫面仍須由圖片工具產生。

獨立教學內容。GPT-6 與 Astra 為 OpenAI 產品,Claude 屬於 Anthropic,Gemini 與 Nano Banana 為 Google 產品;本頁並非官方合作或背書。 2026 年 10 月 15 日更新

Script-to-comic example image with sequential comic panels to illustrate a structured character and panel handoff
圖片來源: LlamaGen script-to-comic gallery visual. Claude Skill produces planning text; this image is not rendered by Claude. 查看來源 ↗

此頁目前僅部分翻譯;詳細實作案例、提示詞及常見問答仍為英文。完成全文翻譯及審核前,不會開放搜尋引擎索引。

重點解答

建立 SKILL.md,定義角色設定、頁面、分格和連續性規則。真正的漫畫畫面仍須由圖片工具產生。

The search intent behind 'Claude comic book skill' is not just another picture generator: creators want a repeatable process that takes a story into scenes, panels, dialogue and production-ready instructions. A custom Claude Skill is a useful place to define that contract.

模型能力重要說明 This tutorial is independent of Anthropic. Claude Skills provide instructions and resources; installing a skill does not automatically add an image-generation capability.

資料實際支持的結論 The location and SKILL.md structure follow the Claude Code Skills documentation. The comic-specific contract is an independent LlamaGen editorial example. Verify the source ↗

Worked example · editable production brief

A concrete Claude Skill folder and output contract

A runnable project skill contains a SKILL.md and optional reference files. The table maps each file or artifact to a practical job, rather than promising that a skill draws pictures.

File / deliverable.claude/skills/comic-book/SKILL.md
ResponsibilityFrontmatter + when-to-use instructions + scene workflow
Acceptance checkClaude recognizes /comic-book
File / deliverablereferences/characters.md
ResponsibilityC01/C02 stable visual traits and dialogue voice
Acceptance checkNo conflicting wardrobe specs
File / deliverableoutput/storyboard.md
ResponsibilityPages > Panels > camera/action/dialogue
Acceptance checkEvery panel has numbered reading order
File / deliverableoutput/render-prompts.md
ResponsibilityOne prompt per panel for an image tool
Acceptance checkVisual directions omit exact dialogue text

Start Claude Code in the project root, invoke /comic-book and test a two-page story before adding automation.

操作教學

Follow the sequence using an original or authorized project. The workshop above is an editorial plan, not output falsely attributed to a model.

  1. Step 1

    Define the exact comic deliverable

    Choose a 4-panel strip, one comic page or a multi-issue outline. Specify reading direction, target audience, language and whether the deliverable is a script, storyboard or finished artwork.

  2. Step 2

    Create the Skill in your project

    Put SKILL.md in .claude/skills/comic-book/. Use YAML frontmatter with a lowercase hyphenated name such as comic-book and a clear description. Store character templates alongside it.

  3. Step 3

    Separate script, layout and visual prompts

    For each panel require purpose, action, composition, dialogue, onomatopoeia and a character-reference ID. Keep page-level pacing separate from image-model prompts.

  4. Step 4

    Review the text plan before rendering

    Check reading order, speech bubble length, recurring props, continuity and whether each page advances the story. Reject vague camera notes.

  5. Step 5

    Render in a dedicated editor

    Use the storyboard and panel prompts in a tool like LlamaGen. Check lettering, balloons, cropping and consistent faces before export.

可複製的提示詞

Replace placeholder details with your own original story and reference assets. Paste into a compatible model or workflow; available tools and results differ.

1. Example SKILL.md

---
name: comic-book
description: Plan a comic script and panel-level storyboard when the user asks for comics, manga or illustrated stories.
---
# Comic Book Workflow
1. Ask for genre, reader age, page count and reading direction.
2. Return a character bible with stable IDs and visual traits.
3. Write pages with numbered panels: shot, action, dialogue, SFX, mood and visual prompt.
4. Preserve names, costumes and setting between panels.
5. End with a continuity checklist and unresolved decisions.
Do not claim images or a PDF were created unless tools actually produced them.

2. Invoke the skill for a short comic

Write a 2-page science-fiction comic for ages 12+. Six panels per page maximum. Begin with a character bible, then provide each panel's camera framing, dialogue, visual prompt and continuity check.

Failure → Cause → Fix

漫畫創作常見問題與修正方法

Specific failure patterns matter more than a generic 'try again'. These fixes target the output you can inspect.

1. The skill is not visible

可能原因 It is outside the recognized project or personal skills directory, or lacks SKILL.md.

處理方式 Move to .claude/skills/comic-book/SKILL.md and restart/check project trust.

2. Claude returns prose instead of panels

可能原因 The schema and panel fields are missing from the Skill.

處理方式 Require page, panel, shot, action, dialogue, SFX and prompt fields explicitly.

3. Visual outputs are claimed but absent

可能原因 A reusable instruction file was confused with an image rendering tool.

處理方式 Keep script deliverables separate; integrate an actual supported image generator.

發佈前檢查項目

Review progress: 0 of 4 checks complete

A personal QA worksheet, not an automated pass/fail evaluation of your image.

常見問題

Is there an official Claude comic-book skill?

Anthropic documents a general custom Agent Skills system. This page teaches you to author a comic-specific skill; it does not claim an Anthropic-official comic skill.

Can a SKILL.md generate images by itself?

No. A skill is reusable guidance; image creation still requires an available tool or application.

How should I handle long books?

Split by scene or issue. Maintain a shared character bible and review each batch for continuity before rendering.

參考來源

Primary documentation and public creator discussions; not a substitute for hands-on results.