Long lecture videos are terrible for quick revision. You cannot Ctrl+F a two-hour recording. So I built a pipeline that samples frames from each video and turns them into ordered PDFs, one per lecture, plus a merged master document I can flip through the night before an exam.

#The problem: you can't Ctrl+F a video

Before exams I did not want to scrub through hours of recorded lectures just to find one slide. Taking screenshots by hand is tedious and easy to lose. So the tool samples frames at a low frame rate with ffmpeg and stitches them into a PDF you can skim, search visually, and annotate.

#How it works

1. 1Recursively scan video/ for supported files (.mp4, .webm, .mkv, .mov, .avi)
2. 2Sort videos by leading number when present (001, 002, ...)
3. 3Extract sampled frames per video with ffmpeg
4. 4Convert the frames into one PDF per video
5. 5Mirror the source folder structure under pdf/
6. 6Optionally merge everything into pdf/final_merged.pdf

#Install and run

pip install pillow pypdf tqdm

Drop your videos into video/ (nested course folders are fine) and run:

python main.py

A few flags make it flexible:

python main.py --fps 0.05 --workers 2 --keep-frames

• ▹--fps controls how many frames ffmpeg samples
• ▹--workers processes several videos in parallel
• ▹--use-gpu enables CUDA acceleration for ffmpeg
• ▹--keep-frames keeps the extracted JPGs after the PDF is built
• ▹--skip-merge skips creating the merged master PDF

#Nested folders stay organised

Point it at a whole course and the output mirrors the input tree. This input:

video/
└── Python Course/
    ├── Module 01/
    │   ├── 001 Intro.mp4
    │   └── 002 Setup.mp4
    └── Module 02/
        └── 001 Variables.mp4

becomes this output:

pdf/
├── Python Course/
│   ├── Module 01/
│   │   ├── 001.pdf
│   │   └── 002.pdf
│   └── Module 02/
│       └── 001.pdf
└── final_merged.pdf

#What I improved along the way

• ▹Renamed the cryptic entry script to a clear main.py
• ▹Added real CLI flags for fps, workers, GPU, frame retention, and merge control
• ▹Validates that ffmpeg actually ran instead of silently continuing on failure
• ▹Auto-creates video/ and pdf/ on first run
• ▹Skips merging final_merged.pdf back into itself on reruns
• ▹Writes clearer logs to run.log when a step fails

It turned a folder of recorded lectures into a searchable stack of PDF notes in one command. The full script and options are on GitHub.