Automated pipeline that turns books into narrated video content for Book!e ("Book!e Inspires Everyone"), a Vietnamese reading community running for 9+ years.
Each book goes through a 7-phase production flow:
- Research — Extract key ideas, angles, and narrative structure from the source book
- Storyboard — Break the narrative into visual sections with scene descriptions
- Script — Write the Vietnamese narration script with timing markers
- Voice — Generate Vietnamese voiceover using viXTTS (self-hosted TTS)
- Visuals — Generate scene images via Google Gemini API, sync to script timing
- Assembly — Compose final video in Remotion with subtitles, BGM, and visual overlays
- Post-production — Generate YouTube/Facebook metadata, catalog insights
| Interface | For | Entry Point |
|---|---|---|
| CLI | Power users | make produce BOOK=<slug> + Claude Code skills |
| n8n Web UI | Visual workflow | Form-based triggers at localhost:5678 |
| Google Sheets | Non-technical users | Add row → pipeline auto-runs |
| Tool | Purpose |
|---|---|
| Remotion 4.0 | React-based video renderer (composition, subtitles, overlays) |
| viXTTS | Vietnamese text-to-speech, self-hosted |
| Google Gemini API | Scene image generation |
| Claude API | Script writing, storyboard, research, metadata |
| n8n | Workflow orchestration (Docker, 10 workflows) |
| Google Sheets | Pipeline control panel (Sheet Orchestrator) |
| Make + bash | Pipeline automation |
| Host bridge | HTTP relay for n8n → host command execution |
| Git LFS | Media file storage (voiceovers, rendered videos) |
Bookie/
├── projects/ai-book-video/
│ ├── Makefile # CLI pipeline entry point
│ ├── WORKFLOW.md # Full pipeline documentation
│ ├── books/ # Per-book data (notes, scripts, audio, video)
│ ├── scripts/ # Shell scripts (voice, images, render, etc.)
│ ├── remotion/ # React video renderer
│ ├── knowledge-base/ # Cross-book intelligence vault
│ ├── brand/ # Style guide, voice reference
│ ├── templates/ # Narrative templates
│ └── n8n/ # n8n orchestration layer
│ ├── docker-compose.yml
│ ├── workflows/ # 10 workflow JSONs (importable)
│ ├── prompts/ # LLM prompts extracted from skills
│ ├── host-bridge/ # HTTP server for host command execution
│ └── README.md # n8n setup guide
├── .claude/skills/ # Claude Code skill definitions
└── .memory-bank/ # Project knowledge
# Production pipeline (voice → images → subtitles → render)
make produce BOOK=atomic-habits
# Or step by step with Claude Code skills
/extract-notes atomic-habits # Research + angles
/create-storyboard atomic-habits # Story direction
/write-video atomic-habits # Script as TTS chunks
make voice BOOK=atomic-habits # Generate voiceover
make render BOOK=atomic-habits # Render videocd projects/ai-book-video/n8n
cp .env.example .env # Fill in API keys
docker compose up -d # Start n8n
./init.sh # Auto-create owner account
./host-bridge/bridge.sh & # Start command bridge
# Open http://localhost:5678- Set up n8n (above) + Google Sheets OAuth2 credential
- Activate workflow "00 — Sheet Orchestrator"
- Add book row with
status = ▶ Start→ auto-runs
| # | Book | Slug |
|---|---|---|
| 1 | Atomic Habits — James Clear | atomic-habits |
| 2 | Gia Dinh — Hector Malot | gia-dinh-hector-malot |
| 3 | Sa Mon Khong Hai | sa-mon-khong-hai |
- CLI pipeline — Makefile + bash scripts + Claude Code skills for full book-to-video production
- Remotion video renderer — React compositions, Ken Burns motion, subtitles, BGM, brand overlays
- viXTTS voice synthesis — Self-hosted Vietnamese TTS with pace-aware gaps, benchmarked CPS
- Knowledge Vault — Cross-book intelligence (concepts, authors, connections, contradictions)
- n8n orchestration — 10 workflows (Docker), host bridge for command execution, form-based UI
- Google Sheets control panel — Sheet Orchestrator polls every 1min, auto-runs production phases
- LLM prompt extraction — 6 Claude Code skills converted to standalone prompts for n8n AI Agent
- NotebookLM as content store — Move scripts, research, chunks from local files to NotebookLM. Single source of truth, collaborative access.
- Interactive Sheet UX — Angle selection via dropdown, script review inline, video preview, one-click YouTube publish. Full pipeline from Sheets alone.
- Sheet visual polish — Conditional formatting, progress bars, color-coded phases, branded header.
- LLM quality tuning — Compare n8n AI Agent output vs Claude Code skill output, iterate prompts.
- Error handling & notifications — Error Trigger nodes, Discord/Telegram alerts on failure.
Book!e Inspires Everyone is a Vietnamese reading community founded in 2016. The AI book video pipeline is one arm of the community's content production, turning book insights into accessible video format for Vietnamese readers.
This repo uses Git LFS for media files (~355MB across 9 tracked files — voiceover WAVs, rendered videos, reference audio). Free GitHub LFS has 1GB/month bandwidth. Run git lfs install before cloning.
All rights reserved.