🎬 AI NewsReel Generator

Automated "Text-to-Video" Agentic System using Google Gemini & Wan 2.2.

AI NewsReel is a fully autonomous video generation pipeline that turns any news topic into a professional, 60-second news explainer video suitable for YouTube Shorts, Instagram Reels, or TikTok. It leverages a Multi-Agent Architecture with NVIDIA L4 GPU for cost-efficient INT8 quantized video generation.

🚀 Key Features

• 🤖 Multi-Agent Orchestration: Four specialized AI agents work in sequence:
  • Researcher: Fetches real-time news (Strict "Last 2 Years" filter) from Google News RSS.
  • Scriptwriter: Drafts broadcast-quality narration (Journalistic 5W+1H style) with complete endings.
  • Scene Planner: Segments scripts into a strict 5-scene structure for maximum engagement.
  • Visual Director: Generates accurate AI video prompts using a single-shot Gemini 2.0 call.
• ⚡ Optimized for Free Tier:
  • Smart Quota Handling: Detects 429 Daily Limit errors and stops gracefully.
  • Batch Processing: Visual Planner uses a single API call for all scenes (80% quota reduction).
  • Token Optimization: Increased limits to prevent narration truncation.
• Core: Python 3.10+
• LLM: Google Gemini 2.5 Flash (google-genai)
• Video Gen: Wan 2.2 INT8 (GCP L4 GPU)
• GPU: NVIDIA L4 (24GB VRAM)
• Frontend: Streamlit
• Video Engine: MoviePy
• Audio: Edge-TTS (Neural)
• Config: YAML & Pydantic for rigid type validation

⚙️ How It Works

1. Input: User provides a topic (e.g., "Aravali Protest").
2. Process:
   • NewsCollectionAgent: Scrapes relevant articles using a 2-year recency filter.
   • SummarizationAgent: Creates a factual summary (no 2018 noise).
   • ScriptWriterAgent: Writes a 5-scene script with a definitive conclusion.
   • VisualAssetAgent: Generates videos using Wan 2.2 (or falls back to images).
   • VideoComposerAgent: Stitches everything into a final .mp4.
3. Output: A polished 1080p vertical video file.

📦 Installation

git clone https://github.com/chaudhary-pawan/NewsReel.git
cd NewsReel
python -m venv venv
# Activate venv:
# Windows: venv\Scripts\activate
# Mac/Linux: source venv/bin/activate
pip install -r requirements.txt

Configuration: Create a .env file and add your Keys:

# Gemini (Required for Agents)
GOOGLE_API_KEY=your_gemini_key_here

GCP L4 GPU Setup (Optional):

For cost-efficient cloud video generation, you can use NVIDIA L4 GPU on GCP with INT8 quantization.

Quick setup:

1. Create an L4 Spot instance on GCP
2. Clone the Wan-Video repository
3. Download INT8 model weights to ./weights/Wan2.2-I2V-14B-720P-INT8
4. Update core/config/settings.yaml with your GCP details:

   gcp:
     project_id: "your-project-id"
     zone: "us-central1-a"
     instance_name: "wan-video-l4-vm"
     bucket_name: "your-bucket-name"

Run:

# GUI Mode
streamlit run interface/app.py

# CLI Mode
python main.py --topic "Your Topic Here"

🏗️ Architecture

User Input → News Collection Agent (Recency Filter) → 
Summarization Agent (Journalistic Tone) → 
Script Writer Agent (Complete Endings) → 
Scene Planner Agent (Max 5 Scenes) → 
Visual Asset Agent (Wan 2.2 Video) → 
Voiceover Agent → Video Composer Agent → 
Final Video Output

📁 Project Structure

NewsReel/
├── core/                # Backend Logic (Agents, Workflows)
│   ├── agents/          # Individual agent implementations
│   ├── workflows/       # LangGraph workflow definitions
│   ├── utils/           # Helper functions (WanClient, LLMClient)
│   └── config/          # Configuration files (settings.yaml)
├── docs/                # Documentation
│   └── archive/         # Archived documentation
├── scripts/             # Setup & Verification Scripts
│   └── archive/         # Archived scripts
├── interface/           # Frontend (Streamlit)
│   └── app.py           # Streamlit UI Entry Point
├── tests/               # Verification Scripts
├── output/              # Generated videos
└── main.py              # CLI entry point

📄 License

MIT License - Free for educational and commercial use

👨‍💻 Author

Created by Chaudhary Pawan as a demonstration of modern Agentic AI + GenAI capabilities.