PDF Transcription with Gemini

This project provides a command-line application for transcribing PDFs using Google's Gemini model.
Example output here. It converts PDF pages into images, uploads them to Gemini, and returns transcriptions in Markdown format. Dependencies handled by uv, see pyproject.toml.

Usage

Run the script with the path to the PDF and any additional arguments you need:

python __main__.py --api_key=YOUR_GEMINI_API_KEY path/to/input.pdf \
    --output=transcription.md \
    --page_limit=5

Command-line arguments include:

• --api_key: Your Gemini API key (optional if set in the environment or in a .api_key file).
• --page_limit: Process up to N pages (0 = no limit).
• --output / -o: Output file path (if not provided, prints to stdout).
• --workers_save, --workers_upload, --workers_chat: Number of parallel workers for each stage.
• --no_overlap: Run each stage sequentially instead of pipelined.
• --thinking: Use the “thinking” model variant with a reduced rate limit.

Contributing

If you have any fixes or improvements, feel free to open a pull request or submit an issue.

License

MIT License. See LICENSE file for details.