How to use Automated Video Generator to create AI-generated videos from text scripts.
The agentic pipeline is fully self-contained and requires no API keys:
# Single video from topic (all-in-one)
npm run agentic -- --topic "5 benefits of drinking water" --title "Hydration" --orientation portrait
# With GPU acceleration
npm run agentic -- --topic "AI in healthcare" --gpu
# Square format for Instagram
npm run agentic -- --topic "Morning routine tips" --format square
# Dry-run: preview plan without downloading or rendering
npm run agentic -- --topic "Space exploration" --dry-run
# Landscape video with images (photo slideshow)
npm run agentic -- --topic "Top 10 coding tools" --orientation landscape --images
# High quality, cinematic preset
npm run agentic -- --topic "Deep sea creatures" --quality high --preset cinematic
# Disable background music
npm run agentic -- --topic "Silent film history" --no-sfx
# Remotion renderer (instead of ffmpeg)
npm run agentic -- --topic "Web development trends" --renderer remotion
# With chapters and verbose ffmpeg logging
npm run agentic -- --topic "JavaScript basics" --chapters --verbose
You can also run individual stages of the pipeline:
# 1. Plan — generate the script structure
npm run agentic:plan -- --topic "Climate change explained" --file input/scripts/agentic-scripts.json
# 2. Download visuals
npm run agentic:visuals -- --file input/scripts/agentic-scripts.json
# 3. Generate voiceover
npm run agentic:voice -- --file input/scripts/agentic-scripts.json
# 4. Render the final video
npm run agentic:render -- --file input/scripts/agentic-scripts.json
# 5. Apply edits (transitions, grades)
npm run agentic:edit -- --file input/scripts/agentic-scripts.json
npm run generate -- --script "Your video script here"
npm run devhttp://localhost:3001/Access CC-licensed videos from Wikimedia Commons and Internet Archive without any registration.
Via HTTP API:
# Search
curl "http://localhost:3001/api/free-video/search?keyword=nature&source=all&count=5"
# Download
curl -X POST http://localhost:3001/api/free-video/download \
-H "Content-Type: application/json" \
-d '{"url":"https://...", "title":"nature-clip"}'
Via Frontend:
http://localhost:3001/video-downloadVia MCP:
search_free_video { "keyword": "space", "source": "all", "count": 3 }
download_free_video { "url": "https://...", "title": "clip" }
Pipeline fallback: Free sources activate automatically when Pexels/Pixabay return no results. No config needed.
Docs: FREE_VIDEO.md — SEO/AEO/GEO-optimized reference
Use [Visual: description] tags in your script for exact visual control over each scene.
Generate multiple videos from a topic list:
# Batch generate from topics
npm run agentic:generate -- --topics "AI coding,Video editing,Photography,Web development"
# Batch with preview (plan-only for each)
npm run agentic:generate:preview -- --topics "AI coding,Video editing"
# Run specific batch stage
npm run agentic:mode:plan
npm run agentic:mode:images
npm run agentic:mode:voice-edgetts
npm run agentic:mode:compose
Enable hardware-accelerated rendering with the --gpu flag or GPU_ACCEL=true in .env:
# CLI flag (auto-selects best encoder)
npm run agentic -- --topic "Space" --gpu
# Environment variable
echo "GPU_ACCEL=true" >> .env
Supports AMD (AMF), NVIDIA (NVENC), and Intel (QSV) GPUs. Falls back to software encoding if no compatible GPU is found.
Preview the full pipeline without downloading or rendering:
npm run agentic -- --topic "Quantum computing" --dry-run
Outputs: title, scene count, orientation, duration, per-scene keywords and voiceover text, and asset decisions — without touching any external service or disk.
Connect Claude Code or Claude Desktop:
claude mcp add automated-video-generator -- npx automated-video-generator
Generated videos are saved as MP4 files in workspace/jobs/<jobId>/render/ with:
--chapters is set)Supports portrait (9:16), landscape (16:9), and square (1:1) orientations.