pxt-primetime

Build multimodal AI applications with Pixeltable. Learn data-centric workflows, semantic search, and content generation using video as an example.

Prerequisites

• Python 3.10+
• uv - Fast Python package installer (pip install uv or brew install uv)
• ffmpeg - Required for video processing (brew install ffmpeg on macOS, apt install ffmpeg on Linux)

Setup

1. Clone this repo and navigate to it:

 git clone https://github.com/apreshill/pxt-primetime/tree/main
 cd pxt-primetime

2. Install all dependencies (uv will automatically create a virtual environment):

 uv sync

uv creates a .venv folder and installs all dependencies. If you already have Pixeltable installed, uv sync will upgrade it to the required version (>=0.5.13) along with all other dependencies.

3. Configure API keys:

Gemini API Key (required for Act 3):

• Get your free API key from aistudio.google.com
• See Pixeltable's API key configuration guide for detailed setup instructions
• Set as environment variable:

  export GEMINI_API_KEY='your-api-key-here'

• Or add to ~/.pixeltable/config.toml:

  [gemini]
  api_key = "your-api-key-here"

OpenAI API Key (optional for Act 2):

• Only needed if you want faster transcription using OpenAI's Whisper API instead of the local model
• Set as environment variable:

  export OPENAI_API_KEY='your-api-key-here'

4. Open VS Code/Cursor in this project folder (or reload if already open)

Dependencies

All dependencies are automatically installed when you run uv sync (see Setup section above). This project uses Pixeltable v0.5.15. The following is a reference of what gets installed - you don't need to install these manually:

Core Dependencies

• pixeltable (>=0.5.15): Primary library for this project. Provides table, view, and computed column functionality.

Video Processing

• opencv-python: Required for video processing operations
• scenedetect: Required for scene boundary detection in videos
• pytube: For downloading videos from YouTube (if needed)

Text Processing & Embeddings

• sentence-transformers (>=2.0.0): Required for Hugging Face sentence embeddings
  • Used in: 02_iterating-with-data.ipynb - creating embedding indexes on transcript text

Audio Processing & Transcription

• openai-whisper: Required for local Whisper transcription (pxt.functions.whisper.transcribe())
  • Used in: 02_iterating-with-data.ipynb - transcribing audio chunks
• openai (optional): For OpenAI Whisper API transcription (pxt.functions.openai.transcriptions())
  • Used in: 02_iterating-with-data.ipynb - faster transcription alternative (requires API key)
  • Note: Install separately with uv add openai if you want to use the OpenAI API instead of local Whisper

Video Generation

• google-genai: Required for Gemini video generation (pxt.functions.gemini.generate_videos())
  • Used in: 03_generate-from-data.ipynb - generating intro/outro videos using Gemini's Veo model
  • Requires: Gemini API key (free tier available)

Development

• ipykernel: For running Jupyter notebooks
• jupyter: For notebook interface

Notebooks

This project contains three progressive notebooks that build on each other. Run them in order (Act 1 → Act 2 → Act 3) as each notebook depends on tables and data created in previous notebooks:

• 01_data-centric-workflows.ipynb - Act 1: Data-centric Workflows
• 02_iterating-with-data.ipynb - Act 2: Iterating with Data
• 03_generate-from-data.ipynb - Act 3: Generate New Content from Data

Running Notebooks

1. Open a notebook
2. Click the kernel selector in the top-right corner
3. Select the kernel showing .venv with the full path to this project's .venv/bin/python

• Look for the path that includes /pxt-primetime/.venv/bin/python
• This confirms you're using this project's environment, not another venv or system Pytho