im2txt but for video

[chapmanjacobd]

What is the problem that is being solved with the new feature?

I would like to extract more metadata from videos, objective data that can be used to cluster similar videos together, preferably offline and output <1kb per row per column.

Enumerate an unordered list of alternatives that you've thought about

• extract frames and run im2txt
• foundation model
• use subtitles or generate captions from audio

If applicable, state your a preferred solution

It would be nice if there was an existing C, Rust, or Python application or library that can do this already