Gemini Video Transcript CLI

This repository provides a command-line workflow that converts batches of local video files to audio (.m4a) and submits them to Google’s Gemini 2.5 Flash model for automatic transcription. Use it to bulk-process recordings you already possess—no YouTube links, web UI, or manual uploads required.

Features

• Scans a directory for videos with a specified extension (default mp4).
• Uses ffmpeg to extract audio only when an .m4a copy does not already exist.
• Uploads each audio file to Gemini 2.5 Flash and retrieves a verbatim transcript stored beside the source video.
• Cleans up uploaded assets from the Gemini account to avoid orphaned files.

Requirements

• Python 3.10+
• ffmpeg available on your PATH
• Google Generative AI API access and a GOOGLE_API_KEY
• Python dependencies from requirements.txt (google-generativeai, python-dotenv)

Setup

python -m venv .venv
source .venv/bin/activate  # Windows: .venv\Scripts\activate
pip install -r requirements.txt
touch .env
echo "GOOGLE_API_KEY=your_key" >> .env

Usage

python app.py --dir data/videos --format mp4

• --dir points to the folder with your source files.
• --format is the input video extension (omit the dot). The default is mp4; change to mov, mkv, etc., as needed.
• Each processed video produces:
  • <name>.m4a (audio) — skipped if it already exists.
  • <name>.txt — plain-text transcript emitted next to the video.

Troubleshooting

• ffmpeg missing: install via your package manager (e.g., brew install ffmpeg, choco install ffmpeg) and reopen the shell.
• Authentication errors: ensure .env contains a valid GOOGLE_API_KEY and that the key has access to Gemini 2.5 Flash.
• Quota limits: the script stops when Gemini rejects an upload; retry after confirming usage limits or switch to a paid tier.

Contributing

Issues and pull requests are welcome. Please describe the scenario you processed (--dir, sample formats), list commands you ran, and attach relevant logs or transcript snippets to expedite reviews.