An AI companion that watches videos with you and provides real-time commentary! Nova uses vision AI to see what's on your screen and react naturally like a friend watching alongside you.
📝 Text Mode
- Split-screen interface with live commentary
- Type to chat with Nova about what you're watching
- Text-based interaction
🎤 Voice Mode ⭐ NEW!
- Nova speaks her commentary out loud
- High-quality text-to-speech with 5 voice options
- Voice input - talk to Nova using your microphone
- Compact floating window that stays on top
- Playful and slightly sarcastic
- Uses modern slang naturally (lmao, ngl, fr, etc.)
- Adds emojis when they fit the vibe
- Reacts in real-time to what's happening on screen
- Remembers context during your watch session
- Auto-comments every 45 seconds (configurable)
- Waits 15 seconds after you speak before commenting
- Won't interrupt while speaking
- Anti-hallucination prompts for accurate observations
- Vision AI powered by LLaVA
# Core requirements
pip install ollama pyautogui pillow
# For voice mode
pip install edge-tts pygame
pip install SpeechRecognition pyaudio
# Make sure Ollama is installed and running
ollama pull llava:7b- Clone this repository:
git clone https://github.com/yourusername/watch-party-nova.git
cd watch-party-nova- Install dependencies:
pip install -r requirements.txt- Run Nova:
python watchparty_voice_final.py- Launch the app - Choose Text Mode or Voice Mode
- In Voice Mode: Select Nova's voice (Aria is default - casual & friendly)
- Open a video anywhere on your screen (YouTube, streaming sites, local videos)
- Let Nova watch - She'll automatically comment every 45 seconds
- Interact anytime:
- Text Mode: Type in the input box
- Voice Mode: Press SPACE or click "TALK" button
- SPACE - Push-to-talk (hold while speaking)
- TALK Button - Click to activate voice input
- Status shows what Nova is doing:
- 🎤 Listening... (capturing your voice)
- 🤔 Nova is thinking... (generating response)
- 🎤 Generating voice... (creating audio)
- Speaking (shows what she's saying)
Choose from 5 high-quality female voices:
- Aria - Casual & Friendly (default) ✨
- Jenny - Warm & Conversational
- Sara - Professional but friendly
- Michelle - Expressive
- Ashley - Young & Fun
You can customize Nova's behavior by editing the code:
# Adjust comment frequency (in seconds)
self.comment_cooldown = 45 # Time between auto-comments
# Change how long Nova waits after you speak
time_since_activity >= 15 # Seconds to wait after user input
# Modify personality in prompts
# Look for the prompt strings in get_response() and get_auto_comment()- Screen Capture: Takes screenshots of the left half of your screen
- Vision Analysis: Sends screenshots to LLaVA (vision language model)
- Response Generation: Creates natural, personality-rich commentary
- Voice Synthesis: Uses Microsoft Edge TTS for high-quality speech
- Voice Recognition: Google Speech Recognition for voice input
- LLaVA 7B - Vision language model for understanding video content
- Edge TTS - Neural text-to-speech (en-US-AriaNeural and others)
- Google Speech Recognition - Voice input processing
- Text Mode: ~2-3 seconds response time
- Voice Mode: ~5-10 seconds (includes TTS generation + audio playback)
- Screenshot: Captures left half of screen only
- Memory: ~2GB RAM (LLaVA model)
- OS: Windows, macOS, or Linux
- Python: 3.8 or higher
- RAM: 4GB minimum, 8GB recommended
- GPU: Optional (speeds up LLaVA inference)
- Microphone: Required for voice input
- Internet: Required for voice recognition
pip install edge-tts pygamepip install SpeechRecognition pyaudio
# On Linux, you may also need:
sudo apt-get install portaudio19-dev python3-pyaudio
# On macOS:
brew install portaudioMake sure Ollama is installed and running:
# Install from https://ollama.ai
ollama serve
# In another terminal:
ollama pull llava:7bThe personality prompts are in the get_response() and get_auto_comment() functions. You can adjust them to make Nova more or less expressive.
The window should be 320x420 pixels. If it's still too small:
- Close the app completely
- Restart it
- The new size should apply
Adjust self.comment_cooldown (around line 521):
self.comment_cooldown = 45 # Increase = less frequent, decrease = more frequentEdit the prompts in get_response() and get_auto_comment() to adjust:
- Sarcasm level
- Slang usage
- Emoji frequency
- Commentary style
Edge TTS supports many voices. Add them to self.available_voices:
"New Voice Name": "en-US-VoiceCodeNeural",Edit the geometry line in setup_voice_mode():
self.root.geometry(f"320x420+{x_position}+{y_position}")Modify capture_left_screen() to capture different areas:
# Currently captures left half:
screenshot = pyautogui.screenshot(region=(0, 0, screen_width//2, screen_height))
# Capture right half instead:
screenshot = pyautogui.screenshot(region=(screen_width//2, 0, screen_width//2, screen_height))Contributions are welcome! Some ideas:
- Add more personality options
- Support for other vision models
- Multi-language support
- Video file analysis mode
- Custom voice training
- Persistent memory across sessions
- Export commentary to text file
MIT License - feel free to use and modify!
- Ollama - For easy local LLM deployment
- LLaVA - Vision language model
- Edge TTS - High-quality text-to-speech
- PyAutoGUI - Screen capture
Having issues? Open an issue on GitHub or check the troubleshooting section above!
Made with 🔥 by someone who wanted an AI friend to watch TikToks with
Have fun watching with Nova! 🐿️🎬✨