Glimpsify (ytvideo2pdf)

Glimpsify extracts slide-like frames from educational videos and builds a PDF of the key visuals (diagrams, formulas, charts). It is optimized for lecture-style videos where text appears on screen over time.

Try now (without setup)

Try it out here: https://colab.research.google.com/drive/1xz6uHeY0QAzMTR8DbXJY8BSvNmKhI24Q?usp=sharing

Quick start

1. Install OCR engine (required for text detection)

   • Windows: install Tesseract OCR and make sure tesseract is on PATH.
   • macOS: brew install tesseract
   • Debian/Ubuntu: sudo apt-get install tesseract-ocr

2. Install the package

pip install ytvideo2pdf

3. Run the CLI

ytvideo2pdf --input=youtube --url="https://youtu.be/Z_MLrbI1s2E"

Common usage

Extract from a local folder (expects a single video file in the directory):

ytvideo2pdf --input=local --dir="C:\path\to\video_dir"

Run with a specific extraction strategy:

ytvideo2pdf --input=youtube --url="https://youtu.be/Z_MLrbI1s2E" --extraction=prominent_peaks

Extract a fixed number of frames:

ytvideo2pdf --input=youtube --url="https://youtu.be/Z_MLrbI1s2E" --k=10

Extract frames at explicit timestamps (seconds):

ytvideo2pdf --input=youtube --url="https://youtu.be/Z_MLrbI1s2E" --extraction=timestamps --timestamps="30, 95.5, 120"

What you get

• A PDF file in output/ with the extracted frames.
• A JSON metadata file alongside the PDF (same name, .json).
• Intermediate folders (unless --no-cleanup) for extracted frames and cached objects.

Key features

• Multiple extraction strategies to pick the most informative frames.
• OCR-based signal processing (Tesseract by default).
• Optional caching of processed frames for reuse.
• Optional plots of the OCR signal (for debugging and tuning).

CLI options (summary)

• --input: youtube | local | pickle
• --url: YouTube video or playlist URL (for youtube input)
• --dir: local directory path (for local or pickle input)
• --ocr: tesseract | easy_ocr | paddleocr
• --ocr_approval: phash | pixel_comparison | approve_all | reject_all
• --extraction: prominent_peaks | k_transactions | key_moments | timestamps | rate_change_threshold
• --k: number of frames to extract, or auto
• --timestamps: comma-separated seconds (for timestamps extraction)
• --threshold: integer threshold for rate_change_threshold
• --cache-frames/--no-cache-frames
• --skip-plot/--no-skip-plot
• --cleanup/--no-cleanup

For Python API usage, see LIBRARY.md.