paldom/video-to-skill
GitHubVideo to knowledge base
Paste a YouTube link or drop in an MP4 and get back a markdown note you can actually trust — every claim deep-linked to the second it was said, the moments that only make sense on screen embedded as screenshots.
Video extraction primitives
The two halves of the pipeline, usable on their own: the words, with a quality gate that tells you when a transcript is lying to you, and the pixels, with scene detection tuned to the video in front of it rather than to a number from a blog post.