VideoLM Factory is a full-stack video generation workspace. It combines research ingestion, NotebookLM automation, Gemini-compatible generation, FFmpeg rendering, and a queue-backed API that can be used from the hosted web UI.

Current reviewer URL:

https://54-162-84-165.sslip.io/engine-demo

The project has several moving parts, so reviewers should be able to understand the system from screenshots before running a long render. Add the images below to the root assets directory with the exact file names shown here.

• React + Vite frontend for creating videos, running NotebookLM research, uploading sources, and downloading generated artifacts.
• NestJS backend that serves the app and exposes /api/... endpoints.
• BullMQ + Redis render queue so heavy FFmpeg jobs run one at a time.
• NotebookLM integration through notebooklm-mcp-cli.
• Docker Compose production stack with Caddy HTTPS reverse proxy.
• Public no-auth demo bridge for reviewer smoke tests.

• Architecture
• API Reference
• NotebookLM Setup
• Deployment
• Operations
• Security Notes
• AWS Production Runbook

export APP_URL="https://54-162-84-165.sslip.io"

curl -fsS S "$APP_URL/api/video/demo/health"

Expected response:

{
   "status": "ok",
   "service": "VideoLM Demo Bridge",
   "baseUrl": "https://54-162-84-165.sslip.io",
   "timestamp": "..."
}}

If a client receives HTML or Unexpected token '<'t;', it is calling a frontend route instead of an API route. API clients must call /api/....

Start with the hosted demo instead of local setup:

1. Open https://54-162-84-165.sslip.io/engine-demo.
2. Click [run] $ homes-engine health.
3. Click [run] $ homes-engine demo.
4. Confirm the terminal logs cache, verifying mp4, HEAD 200, content-type: video/mp4, and mp4 verified.
5. Play the MP4 that appears in the page, or open the public MP4 directly:

https://54-162-84-165.sslip.io/videos/videolm_primary_cached.mp4

For advanced NotebookLM artifacts, run $ homes-engine config or $ homes-engine notebooklm --help in the page. The heavy NotebookLM path is intentionally secondary so the reviewer demo is deterministic.

For the full authenticated workspace at https://54-162-84-165.sslip.io, use:

email: [email protected]
password: VideoLMReview2026!

The Register button is optional for new accounts. Reviewers should use the demo credentials above if they only need to inspect the hosted app.

The hosted app avoids local FFmpeg and local environment setup for reviewers. NotebookLM and AI-key setup are documented for users who want to run the full workflow themselves.

Requirements:

• Node.js 20+
• npm
• Redis for queue-backed rendering
• FFmpeg
• Python 3 and uv if using NotebookLM locally

Install dependencies:

npm install

Create a local environment file:

cp .env.example .env

Run frontend and backend together:

npm run dev

Backend only:

cd server
npm install
npm run devev

Install the upstream CLI:

uv tool install notebooklm-mcp-cli

Log in with the browser-based flow:

nlm login
nlm login --check
nlm doctoror

The default profile stores cookies at:

~/.notebooklm-mcp-cli/profiles/default/cookies.json

VideoLM can use that file directly on the server, or users can paste/upload their cookies.json in the Research Lab UI. Full setup details are in docs/NOTEBOOKLM_SETUP.md.

Create .env in the repo root:

GEMINI_API_KEY=...
OPENROUTER_API_KEY=...
HF_TOKENS=...
JWT_SECRET=replace-this
NLM_HOME=/home/ubuntu/.notebooklm-mcp-cli
PUBLIC_BASE_URL=https://54-162-84-165.sslip.ioip.io

Start the stack:

docker compose up -d --build

Expected containers:

videolm_factory
videolm_redis
videolm_caddy
y

Caddy serves HTTPS on port 443 and reverse-proxies to the app container on port 3001.

The no-auth demo bridge is intended for reviewer smoke tests.

GET  /api/video/demo/health
POST /api/video/demo/assemble
GET  /api/video/:projectId/status
GET  /videos/*.mp4
p4

POST /api/video/demo/assemble expects multipart/form-data:

audio      required WAV/MP3 narration file
images     required JPG/PNG/WebP scene files, up to 100
script     optional script text for captions
bgMusic    optional background music upload
bgMusicId  optional file name from server/data/music
projectId  optional stable id for polling
ling

Use docs/API.md for exact request and response shapes.

npm run build

cd server &∓& npm test -- --runInBand
curl -fsS https://54-162-84-165.sslip.io/api/video/demo/healthh

App.tsx                         Main React shell
components/ResearchDashboard.tsx NotebookLM research UI
services/ffmpegService.ts        Browser-side video assembly client
server/src/research              NotebookLM research API
server/src/video                 FFmpeg render API and queue bridge
server/public/videos             Generated video files
server/data/database.sqlite      SQLite database
docker-compose.yml               Production services
Caddyfile                        HTTPS reverse proxy
e proxy

This stack is CPU-bound and can run without a GPU. For a short review window, the main AWS costs are EC2 instance hours, EBS disk, snapshots, data transfer, and public IPv4. Stop unrelated VMs, keep one production VM online, and set a billing alert.