AI-powered universal video pipeline Give an AI Agent a playbook, and it will forge any insight into a compelling short video.
FableForge is a fully automated video production pipeline that turns any concept into a polished short video — complete with narration in your cloned voice, AI-generated visuals, frame-accurate subtitles, and cinematic Ken Burns motion.
FableForge: A universal AI video production pipeline.
| Title | Management Insight | Duration | Link |
|---|---|---|---|
| The Scale That Gains Weight | Cantillon Effect & wealth redistribution | 133s | Watch on YouTube |
| The Lighthouse of Fogport | Time Inconsistency & The cost of broken promises | 82s | Watch on YouTube |
📺 The above works are fully generated and rendered using this FableForge SOP with cloned personal voices.
FableForge is packaged as an AI Agent Skill. You don't need to manually clone repositories or install dependencies. Just give the Skill to your AI assistant (like Cursor, Cline, or Gemini), and it will build the studio for you.
Step 1 — Download the Skill In an empty directory, run:
npx skills add Lucas-Kay8/ai-video-studioStep 2 — Tell your AI to initialize Open your AI chat and say:
"Initialize the FableForge studio."
The AI will read the SKILL.md, automatically scaffold the template/ and voice-model/ directories, and download FFmpeg for you.
Step 3 — Record your voice sample
Once the AI finishes initialization, it will prompt you to record a 15-second voice sample and place it in voice-model/01_samples/my_voice.wav.
Step 4 — Generate your first video Tell your AI:
"Help me make an analytical video about [your topic]." The AI will handle the rest of the 5-stage pipeline!
your-workspace/
├── .agents/skills/fableforge/
│ ├── SKILL.md ← AI Agent SOP (Core instructions)
│ └── resources/ ← Templates & Voice models used by AI for init
├── template/ ← Auto-copied by AI during init
├── voice-model/ ← Auto-copied by AI during init
├── bin/ ← FFmpeg (Auto-downloaded by AI)
└── YYYYMMDD/ ← Per-episode archive (Auto-generated by AI)
The heart of FableForge is a command-level executable SOP for AI Agents, stored in .agents/skills/fableforge/SKILL.en.md.
graph TD
classDef stage fill:#1e1e2e,stroke:#3b4252,stroke-width:1px,color:#d8dee9;
classDef active fill:#2e3440,stroke:#88c0d0,stroke-width:2px,color:#88c0d0;
Concept["💡 Stage 1: Concept & Script<br/>(AI self-check & manual gate)"]:::stage
Creative["🎨 Stage 1.4-1.6: Creative Assets<br/>(16:9 Wide Shot + Voice Cloning)"]:::stage
BGM["🎵 Stage 1.7: Music Matching<br/>(Mood awareness & auto-ducking BGM)"]:::stage
Whisper["⏱️ Stage 2: Precision Timeline<br/>(Whisper word-level transcription)"]:::stage
Static["📐 Stage 3: Static Layout Build<br/>(100% no-crop CSS safe zone)"]:::stage
inspect["🎬 Stage 4: Pre-flight Gate<br/>(inspect verification & StaticGuard)"]:::stage
Render["🚀 Final Render: Export MP4 via HyperFrames"]:::active
Concept --> Creative
Creative --> BGM
BGM --> Whisper
Whisper --> Static
Static --> inspect
inspect --> Render
| Stage | What Happens | Exit Criteria |
|-------|-------------|---------------| | Stage 1 Concept & Asset Generation | Write allegory, generate images, synthesize voice | Image count == scene count, audio file ready | | Stage 1.6 Typography Poster | Generate click-optimized pure-text cover and CTA end card | Cover & End card ready | | Stage 1.7 BGM Matching | Mood analysis, track selection, auto-integration | BGM file ready, attribution added | | Stage 2 Data-Driven Timeline | Whisper transcription, frame-accurate scene alignment | Deviation < 0.2s, zero estimated values | | Stage 3 Static Layout Validation | Pure HTML/CSS, verify no image cropping before animation | All images display fully, DOM injected dynamically | | Stage 4 Pre-flight & Render | inspect → render, machine validation replaces eyeballing | inspect exits 0, duration matches audio exactly |
FableForge automatically generates high-impact, pure-text cover posters (scene_cover.png) and Call-To-Action end cards (scene_end.png) for your videos.
- 10 Built-in Styles: From "Modern Tech Blue" to "Minimalist Black & White".
- No Images Rule: Typography-only designs for maximum click-through rates.
- Auto-personalized: The AI will ask for your preferred signature/IP name before generating.
FableForge now supports automatic background music matching:
- Mood-Aware: Matches BGM based on the emotional gear of your script (e.g., The Scale That Gains Weight uses 'Undertow' by Scott Buckley for its atmospheric tension).
- Auto-Ducking: Defaults to 0.15–0.25 volume to ensure narration clarity.
- Attribution Ready: Automatically inserts license info into your script.
| Component | Role | Link |
|---|---|---|
| HyperFrames | HTML-to-MP4 deterministic render engine | hyperframes.heygen.com |
| VoxCPM2 | Personal voice cloning & TTS synthesis | PyPI |
| Whisper | Word-level timestamp transcription | github.com/openai/whisper |
| GSAP | Ken Burns zoom & motion choreography | gsap.com |
| FFmpeg | Audio duration analysis & silence detection | ffmpeg.org |
- Open
template/script-template.mdand fill in your storyboard - Generate images with Midjourney / Flux / DALL·E → name them
scene1.png...scene{N}.png→ drop intoassets/ - Generate
narration.wavwith any TTS tool → drop intoassets/ - Run
npm run check && npm run render
Inject .agents/skills/fableforge/SKILL.en.md into your Agent context. It will autonomously follow the 5-stage pipeline.
Compatible with: Claude / Gemini / Cursor / VS Code Copilot
| Project | Role in FableForge |
|---|---|
| HyperFrames by HeyGen | Core video render pipeline |
| VoxCPM2 | Voice cloning engine |
| OpenAI Whisper | Audio transcription (MIT License) |
PRs welcome for:
- New allegory video examples (add to a
YYYYMMDD/directory) - Additional visual style CSS templates
- TTS / Whisper adapters for other languages
FableForge 是一个全自动通用视频生产管线，能将任何概念或洞察转化为精致的短视频——包含克隆你的声音进行的旁白、AI 生成的视觉画面、精确到帧的字幕，以及电影级的 Ken Burns 动态效果。
FableForge 通用制片厂: 利用 AI 将深刻的洞察铸造成引人入胜的短视频。
| 标题 | 管理洞察 | 时长 | 链接 |
|---|---|---|---|
| 《会自己变重的秤》 | 坎蒂隆效应与财富再分配 | 133s | 在 YouTube 观看 |
| 《雾港城的灯塔》 | 时间不一致性与承诺的代价 | 82s | 在 YouTube 观看 |
📺 以上代表作品均完全基于本 FableForge 工业化流水线 SOP 生成，并使用了克隆的个人声纹进行配音渲染。
FableForge 已经被封装为一个 AI Agent Skill。你不再需要手动 git clone 或敲命令安装依赖。只需将这个 Skill 交给你的 AI 助手（如 Cursor、Cline 或 Gemini），它会自动为你搭建整个制片厂。
第 1 步：下载 Skill 在一个空的文件夹中运行：
npx skills add Lucas-Kay8/ai-video-studio第 2 步：让 AI 初始化环境 打开 AI 对话框，发送：
"请帮我初始化 FableForge 制片厂。"
AI 会自动读取 SKILL.md，为你一键释放 template/ 和 voice-model/ 脚手架，并自动下载配置 FFmpeg。
第 3 步：录制声音样本
初始化完成后，AI 会提示你录制一段 15 秒的声音样本，放在 voice-model/01_samples/my_voice.wav。
第 4 步：生成你的第一个视频 对 AI 说：
"帮我做一个关于[某个话题]的分析类视频。" AI 会自动接管剩下的 5 步工业化流水线！
your-workspace/
├── .agents/skills/fableforge/
│ ├── SKILL.md ← AI Agent 专用 SOP (核心规则)
│ └── resources/ ← 模板与模型资产 (AI 初始化时释放)
├── template/ ← AI 自动释放的视频模板
├── voice-model/ ← AI 自动释放的语音克隆脚手架
├── bin/ ← AI 自动下载的 FFmpeg 依赖
└── YYYYMMDD/ ← 每期视频的项目归档
本项目的核心是一套写给 AI Agent 的命令级可执行 SOP，存放于 .agents/skills/fableforge/SKILL.md.
graph TD
classDef stage fill:#1e1e2e,stroke:#3b4252,stroke-width:1px,color:#d8dee9;
classDef active fill:#2e3440,stroke:#88c0d0,stroke-width:2px,color:#88c0d0;
Concept["💡 Stage 1: 概念策划 & 剧本生成<br/>(AI 概念自检 & 停机确认)"]:::stage
Creative["🎨 Stage 1.4-1.6: 声画资产筹备<br/>(DALL-E 3 16:9生图 + VoxCPM2 声纹克隆)"]:::stage
BGM["🎵 Stage 1.7: 自动化配乐匹配<br/>(情绪档位识别 & 智能音量避让)"]:::stage
Whisper["⏱️ Stage 2: 帧级对齐时间轴<br/>(Whisper 词级断句 -> 绝对时长映射)"]:::stage
Static["📐 Stage 3: 静态网页排版<br/>(DOM 动态注入 & CSS 100% 不裁切安全区布局)"]:::stage
inspect["🎬 Stage 4: 自动化发布预检<br/>(inspect 校验 & StaticGuard 门禁检测)"]:::stage
Render["🚀 最终渲染: HyperFrames 编译导出 MP4"]:::active
Concept --> Creative
Creative --> BGM
BGM --> Whisper
Whisper --> Static
Static --> inspect
inspect --> Render
| 阶段 | 做什么 | 退出标准 |
|------|--------|---------| | Stage 1 概念与资产生成 | 创作内容、生成分镜素材、合成语音 | 主体视觉与音频文件就位 | | Stage 1.6 纯文字海报生成 | 自动生成高转化率纯文字封面与封底 | 封面与封底就位 | | Stage 1.7 BGM 自动配乐 | 情绪识别、曲库匹配、自动下载集成 | BGM 就位，署名信息补充 | | Stage 2 数据驱动时间轴 | Whisper 转录，精确对齐每幕时间 | 误差 < 0.2 秒，无估算值 | | Stage 3 静态排版验收 | 纯静态 HTML/CSS，验证不裁切 | 所有图片完整显示，DOM 动态注入 | | Stage 4 预检与渲染 | inspect → render，机器校验 | inspect 0 报错，时长精确匹配 |
FableForge 会为你的每一个视频自动生成高转化率的封面与封底：
- 10 大排版风格矩阵：涵盖“高级商务风”、“科技蓝”、“黑金战略”等，适配各种内容体裁。
- 纯文字铁律：抛弃配图，采用满屏大字错位排版，最大化视觉冲击力。
- 彻底去个人化：AI 会在生成前询问你的 IP 署名，为你量身定制专属落款。
FableForge 现已支持自动配乐：
- 情绪匹配：根据脚本的“情绪档位”自动推荐合适的背景音乐（如《会自己变重的秤》选用了 Scott Buckley 的 'Undertow' 来烘托悬疑感）。
- 自动避让：默认音量设为 0.15-0.25，确保背景音不掩盖人声旁白。
- 版权合规：自动在脚本中注入作者署名和 CC 协议信息。
| 组件 | 用途 |
|---|---|
| HyperFrames | 视频渲染引擎 |
| VoxCPM2 | 声纹克隆与语音合成 |
| Whisper | 词级时间戳音频转录 |
| GSAP | 动效编排 |
| FFmpeg | 媒体处理 |
MIT