yt-slidesnap — Turn a Lecture Video Into a PDF of Its Slides

What it is

yt-slidesnap takes a lecture video and the timestamps where each question is on screen, and gives back a clean PDF of the slides — one page each, no answers, no video downloaded. It pulls frames straight from the stream, so a 2.5-hour class becomes a tidy document in seconds.

The local web UI: paste a video URL and your timestamps, click once, get a PDF.

The tedious case

It is built for the exact situation that made me write it: a long recorded class with dozens of questions in it, where you want the question slides in a document instead of scrubbing a timeline by hand and screenshotting each one. Doing that once is tedious; doing it every week is a waste of an evening.

How it works

1. You give it a video URL and a list of timestamps.
2. It resolves the stream (no full download) and seeks to each window.
3. For each slide it samples several candidate frames and picks the one where the presenter is least likely to be blocking the slide.
4. It assembles the chosen frames into a single PDF, and writes an auto_picked.png contact sheet so you can eyeball what it chose.

A full run of 16 slides takes roughly 25 seconds, including resolving the stream URL. You do not need ffmpeg installed — a build ships inside the tool's dependencies.

Timestamp handling

The input is deliberately forgiving. Paste whatever you have — Q22: Question 01:34:21 | Answer 01:34:53 (C), a markdown table, or a pipe- or comma-separated list — and it sorts it out.

Two rules exist because guessing wrong would silently capture the wrong frame:

• The first stamp on a line is when the slide appears. Every later stamp is a ceiling and the earliest one wins, so an answer can never leak into the PDF.
• Two rules on ambiguity: a two-part stamp is always minutes:seconds (write 1:34:21 for an hour), and a lone integer is treated as seconds only when it cannot be an id. When in doubt, write 5661s.

Run it with --dry-run first and it prints exactly the window it understood for each row — so a misread shows up before you spend a run on it.

The web UI

If you would rather click than type flags, yt-slidesnap serve starts a local web page on 127.0.0.1:8765 — no account, no cloud. Same engine, a couple of text boxes and a button.

Install it

Install it as a command with pipx install git+https://github.com/pavnxet/yt-slidesnap, or clone the repo and run python slidesnap.py.