Narrator-Camera: Real-Time Personality-Powered Webcam Narrator 🕶️🎤

Narrator-Camera or NarrateCam is a real-time computer vision project that captures webcam footage, uses a large language model (LLM) to generate funny Deadpool-style (or any personality of choice) descriptions of the scene, enhances them with a personality-aware context engine, and finally speaks them out loud using text-to-speech.

🔧 Features

• 📸 Captures live video feed from webcam
• 🧠 Describes frames using an image LLM (LLaVA)
• 🗣️ Adds contextual, personality-driven humor (Deadpool-style)
• 🔊 Speaks the description using text-to-speech (TTS)

🗂️ Project Structure

.
├── main.py               # Core script to capture webcam, send frames, display, and speak
├── llm_api.py            # Handles interaction with image and text LLMs
└── text_to_speech.py     # Converts generated text to speech using pyttsx3

🛠️ Requirements

• Python 3.8+
• OpenCV
• pyttsx3
• OpenAI-compatible LLM server (like LM Studio)
• Local LLaVA (llava-v1.5-7b) and Dolphin (dolphin3.0-llama3.1-8b) models running via http://localhost:1234/v1

Install dependencies

pip install opencv-python pyttsx3 openai

🚀 Getting Started

1. Run your LLMs: Launch your LLaVA and Dolphin models using LM Studio or another OpenAI-compatible server on localhost:1234.

2. Start the app: Run the main script:

   python main.py

3. See it in action:

   • A webcam window opens
   • Every 5 seconds, the current frame is described in Deadpool-style humor
   • The description is read aloud using TTS
   • Press q to quit

🧠 How It Works

1. main.py captures frames from the webcam.
2. Every 5 seconds, the current frame is base64-encoded and sent to the image model (llava-v1.5-7b) for visual description.
3. That description is passed into a contextual personality engine (dolphin3.0-llama3.1-8b) that keeps a brief context history.
4. The final personalized message is spoken aloud using pyttsx3.

🗣 Example Output

🎥 Scene: "A guy sitting in front of a laptop with a look of existential dread. Classic."
🧠 Contextual Deadpool-style message: "Oh, the face of someone trying to debug on a Friday night. Bravo!"
🔊 Spoken via TTS.

📌 Notes

• Make sure LM Studio or the API server is running with the correct models.
• If audio playback fails, check that your system supports pyttsx3 (Windows and macOS generally work out of the box).

📄 License

MIT License