Generate video advertisements from product pages using Claude Agent SDK and AI video generation APIs.

Examples of AI-generated video advertisements (click to play):

Apple Vision Pro	Pioneer DDJ-XP2	Patagonia R1 Fleece
Hermes Scarf	Sony Alpha 7 V	BMW i8

View all generated videos at the Video Showcase when running locally.

This tool takes a product detail page URL as input and automatically:

1. Extracts product metadata (title, description, images, price, brand)
2. Uses Claude to craft a compelling ad script and video prompt
3. Generates a short video advertisement using AI video generation

• Python 3.10+
• Claude Code CLI (for authentication)
• Kie.ai API key (for Veo 3 - default model)
• FreePik API key (optional, for --freepik flag)

1. Clone the repository:

git clone <repo-url>
cd agentic-orchestration

2. Create and activate a virtual environment:

python3 -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

3. Install dependencies:

pip install -e .

4. Set up environment variables:

cp .env.example .env
# Edit .env and add your API keys

5. Authenticate Claude Code (if not already):

claude
# Follow the prompts to authenticate

Start the web server to use the browser-based interface:

ad-generator-server

This starts a FastAPI server at http://localhost:8000 with:

• Web UI for submitting product URLs and tracking progress
• REST API at /api/generate, /api/status/{job_id}, /api/jobs
• Video showcase at /video-showcase.html

# Basic usage (Veo 3 Fast - default, 5s video)
ad-generator "https://example.com/product/123"

# With Veo 3 Quality (higher quality, slower, more expensive)
ad-generator "https://example.com/product/123" --veo3-quality

# Also generate with FreePik WAN 2.6
ad-generator "https://example.com/product/123" --freepik

# Specify duration (5, 8, 10, or 15 seconds)
ad-generator "https://example.com/product/123" -d 10

# Specify output directory
ad-generator "https://example.com/product/123" --output ./my-ads

# Force regeneration (skip duplicate check)
ad-generator "https://example.com/product/123" --force

Each generation creates a timestamped directory with the product name:

output/
└── macbook-air-13-inch-m4-2025-20260116-132438/
    ├── README.md                    # Generation details and prompt
    ├── freepik_<task-id>.mp4        # FreePik WAN 2.6 video
    └── veo3_<task-id>.mp4           # Veo 3 video (if --veo3)

The README.md contains:

• Input product URL
• Product information
• Video generation prompt
• Links to generated videos

The tool automatically detects if a video has already been generated for a URL by scanning existing README.md files. Use --force to regenerate.

src/ad_generator/
├── __init__.py
├── main.py               # CLI entry point
├── agent.py              # Claude Agent orchestration
├── metadata_extractor.py # Product page scraping
├── freepik_client.py     # FreePik API client
├── kie_client.py         # Kie.ai API client (Veo 3)
└── models.py             # Pydantic data models

1. Metadata Extraction: Fetches the product page and extracts structured data using:

   • JSON-LD (schema.org Product)
   • Open Graph meta tags
   • Standard HTML elements

2. Claude Agent: An AI agent analyzes the product and creates:

   • Ad concept and script
   • Optimized video generation prompt

3. Video Generation: The prompt is sent to configured video generation APIs

Install dev dependencies:

pip install -e ".[dev]"

Run tests:

pytest

For production deployment to Vercel, see VERCEL.md for step-by-step instructions including:

• Vercel account setup
• Environment variable configuration
• Deployment process
• Testing and troubleshooting