Well, with things like this, I feel like someone could get interesting results pretty quickly:

https://lintangwisesa.github.io/MediaPipe-in-JavaScript/inde...

https://github.com/cosyneco/MediaPipe.NET

https://ai.google.dev/edge/mediapipe/solutions/guide

I'd think you'd start with a tool to automatically cut video into scenes, and kind of go from there...