沐阳手绘视频

从一个主题，自动生成完整手绘风格动画讲解视频：文案、分镜、AI 图片、自然配音、同步字幕、Remotion 动效、MP4 渲染和成片验证。

Features

• 一句话生成：只输入主题即可默认完成脚本、画面、配音、字幕和视频渲染。
• 真人感旁白：支持 Fish Audio 和 ElevenLabs；中文项目默认优先选择中文/普通话声音，不用外国角色硬配中文。
• 统一配音：默认生成一条连续旁白音频，避免每个镜头声音不连贯。
• 图片生成画面：每个场景使用 imagegen 生成独立手绘插画，字幕和标题由 Remotion 渲染。
• 十种视觉风格：编辑手绘、现代卡片、笔记本、墨稿海报、水彩绘本、卷轴、复古报刊、黑板、蓝图、贴纸。
• 多套排版和动效：默认避免左右分栏，采用更适合短视频阅读的上方标题、中心画面、底部字幕结构。
• 可编辑工程：输出 Remotion 项目，可继续改脚本、图片、字幕、动画和字体。

中文快速开始

只需说：使用 $沐阳手绘视频，主题：为什么存钱越早越轻松。未指定的内容会自动采用推荐值：中文、60–90 秒、竖屏、暖色编辑手绘、逐幕生成图片、同一个自然中文声音的连续旁白、同步字幕和明显但克制的标题/画面动效。若想参与选择，说“引导模式”，skill 会分轮引导风格、画幅、时长和声音；随时说“其余默认”即可停止提问并继续制作。

What users can say

Use $沐阳手绘视频. Topic: Why saving early feels easier.

That is enough. The default run produces a 60–90 second vertical video with Chinese narration, Chinese subtitles, a warm editorial hand-drawn style, generated scene illustrations, one coherent continuous narrator voice, and visible Remotion motion.

For guided creation:

Use $沐阳手绘视频 in guided mode. Help me choose the style and voice.

Requirements

• Codex with this skill installed
• Node.js and npm
• Python 3.10+
• FFmpeg and FFprobe
• Remotion dependencies installed in the generated project
• Access to the imagegen skill
• A Fish Audio or ElevenLabs account and API key for narration

The skill never stores account keys, personal voice IDs, or reference recordings inside its own directory.

Installation

Copy the whole skill folder into your Codex skills directory, normally %USERPROFILE%\.codex\skills\ on Windows or ~/.codex/skills/ on macOS/Linux, then restart Codex. The visible skill name is 沐阳手绘视频; the existing folder may remain create-handdrawn-explainer for compatibility with older projects and scripts.

If you publish this repository to GitHub, keep the folder contents as the repository root:

.
├─ SKILL.md
├─ README.md
├─ agents/
├─ assets/
├─ references/
├─ scripts/
└─ tests/

Fish Audio setup

Fish Audio is a Chinese-first cloud TTS option and is useful when ElevenLabs has no usable Chinese voice in the current account.

1. Sign in at .

2. Create an API key.

3. Create or update .env in the generated project:

   FISH_API_KEY=
   FISH_MODEL=s2.1-pro-free
   # Optional: FISH_REFERENCE_ID=
   

Run. By default this creates one continuous narration file, which keeps the voice identity, pacing, and emotion coherent across the whole video:

python scripts/fish_audio_project.py --project video

For debugging only, you can force one file per scene:

python scripts/fish_audio_project.py --project video --mode segmented

To make a short test sample:

python scripts/fish_audio_project.py --project video --sample --sample-text "你好，我们用十秒钟试一下中文旁白效果。"

ElevenLabs setup

1. Sign in at .

2. Create a restricted key. Enable only Text to Speech and Voices: Read. A small usage cap is recommended.

3. Create .env in the generated project:

   ELEVENLABS_API_KEY=
   

Do not paste the key into chat or commit .env. If Chrome control is available, ask Codex to configure the key from your already signed-in browser; the skill instructs it to save the key locally without displaying it.

Automatic voice selection

For Chinese projects, prefer Fish Audio or a true Chinese/Mandarin ElevenLabs voice. Do not use English role voices for Chinese narration unless the user explicitly requests cross-language casting.

ElevenLabs

The bundled selector ranks voices available to the account:

1. Mandarin/Chinese voices intended for narration or education
2. Warm, professional, informative, or storyteller voices inside that Chinese/Mandarin pool
3. Voices usable by the current subscription

For Chinese narration, the selector does not fall back to an English role voice such as Bella, River, or another foreign-character voice. If every Chinese/Mandarin voice is unavailable on the current plan, generation stops and asks the user to add, enable, or choose a usable Chinese voice. Cross-language casting requires the explicit --allow-cross-language-voice flag.

Visual styles

The default is warm-editorial. Guided mode can choose among editorial, magazine grid, notebook, ink poster, watercolor storybook, calligraphy scroll, retro newspaper, chalk classroom, technical blueprint, and playful sticker systems. See references/style-catalog.md for fonts, palettes, and use cases.

Style gallery

Each preset controls the image prompt direction, palette, heading font mood, caption font stack, background texture, and default layout family.

The ten built-in style IDs are:

Text and image layout gallery

The Remotion template keeps generated images text-free, then renders all titles, scene numbers, captions, frames, and subtitles in code. The current presets use ten layout families instead of one repeated frame: editorial opener, grid spotlight, notebook diary, ink poster, storybook plate, scroll centerpiece, newspaper poster, chalk lesson, blueprint callout, and sticker stage. Default presets avoid left-right split layouts; title, image, and subtitle follow a top-center-bottom reading flow.

Typography range

The style catalog includes multiple font moods for guided mode: serif editorial, humanist sans, handwritten note, brush poster, soft literary, calligraphy title, newspaper serif, chalk marker, technical grotesk, and rounded playful. Decorative fonts are used only for short headings; captions stay on readable sans or serif stacks.

Motion presets

The Remotion template includes reusable motion presets. If motion.id is omitted, the selected visual style automatically chooses its matching motion language. Every default scene animates the heading, focal image, subtitle bar, texture/accent layer, and image reveal/drift. Subtitles are visible by default and are rendered by Remotion, not baked into generated images.

Manual command flow

python scripts/preflight.py --project video
python scripts/create_project.py --plan plan.json --output video
# Use Codex + imagegen to create every public/images/.png listed in plan.json.
python scripts/fish_audio_project.py --project video
# or, if the account has a true Chinese/Mandarin voice:
python scripts/elevenlabs_project.py --project video --auto-voice
cd video
npm install
npm run render
python ../scripts/verify_output.py out/video.mp4 --plan src/generated/plan.json

To list voices:

python scripts/elevenlabs_project.py --project video --list-voices

To force a selected voice:

python scripts/elevenlabs_project.py --project video --voice "Voice name"

To create three comparable guided-mode samples:

python scripts/elevenlabs_project.py --project video --sample-voices --sample-text "你好，我们用一分钟，把这件事讲清楚。"

Output

Each generated project is editable and contains:

video/
├─ public/images/          generated scene art
├─ public/audio/           continuous narration.mp3 by default
├─ script/                 readable exported scripts
│  ├─ voiceover-script.md  full narration script
│  ├─ captions-script.md   on-screen caption script
│  └─ scene-plan.md        scene titles, image prompts, captions, narration
├─ src/generated/plan.json
├─ src/generated/captions.json
├─ src/stylePresets.ts
└─ out/video.mp4

Troubleshooting

• Missing ELEVENLABS_API_KEY: finish the setup above; do not use SAPI as a substitute.
• Missing FISH_API_KEY: create a Fish Audio key and save it in the project .env.
• Fish Audio 401: the key is missing or invalid.
• Fish Audio 402: the account has insufficient credit or selected model is unavailable.
• 402 paid_plan_required: choose a premade voice or run --auto-voice to fall back automatically.
• No usable Chinese/Mandarin voice: add or enable a Chinese voice in ElevenLabs, choose a Chinese voice by name or ID, or explicitly allow cross-language casting.
• 429 concurrent_limit_exceeded: retry sequentially; the bundled generator is sequential by default.
• Missing image: generate public/images/.png before rendering.
• Missing font: the presets contain safe fallbacks; install an OFL font from the style catalog for the intended appearance.
• Voice changes between scenes: rerun Fish Audio without --mode segmented; the default continuous mode creates one unified narrator track.
• Missing subtitles: rerun the TTS project script so it rebuilds src/generated/captions.json, then render again.
• Video cuts narration: rerun the TTS project script, which measures generated audio and rebuilds scene durations.

Open-source release checklist

Before pushing to GitHub:

• Confirm .env, generated projects, audio files, videos, temporary preview projects, and API keys are not committed.
• Confirm assets/previews/style-gallery-contact-sheet.png and assets/previews/layout-gallery-contact-sheet.png render correctly on GitHub.
• Run a representative template type check from a generated project: npx tsc --noEmit.
• Run the skill validator. Note: this skill intentionally uses the Chinese display name 沐阳手绘视频; Codex’s default validator may warn that the name is not hyphen-case.
• Include LICENSE and keep third-party service credentials out of the repository.

License

MIT. See LICENSE.