marcelpanse/youtube-guitar-tab-parser

This is a command line tool that takes a YouTube guitar lesson video and turns the guitar tab sheet music shown on screen into a downloadable PDF. Many guitar lesson videos display scrolling tab notation on screen while the instructor plays, and this tool automates the tedious process of manually screenshotting each new line of that notation as it appears. It works by downloading the video, taking screenshots at regular intervals, and using Claude's image understanding ability to figure out where on screen the tab notation actually appears. Once it knows that region, it crops every screenshot down to just that area. Because the video keeps showing the same line of tab for several seconds while it plays, the tool reads the small bar or measure number printed on each line and keeps only one screenshot per distinct number, throwing away the repeated near identical frames and anything that is not actually sheet music, like an intro or title card. The remaining unique lines are then stacked one after another into a single PDF, with the video's own title used as both the file name and the heading on the first page. Running it requires Node.js version 20 or newer, plus two external command line tools, yt-dlp for downloading YouTube videos and ffmpeg for extracting frames from them, both of which need to already be installed and available on the system. It also requires a paid Anthropic API key, since detecting the tab region and reading the bar numbers both rely on calls to Claude's vision model. Several options let you adjust things like how often frames are sampled, which Claude model is used, and how aggressively near duplicate frames are filtered out before they reach the more expensive vision step, though the defaults are meant to work without any tuning for most videos. The readme does not state a specific software license for the code.