There is significant existing academic work on generating a "story" from those elements as well using various approaches - e.g. the papers from a search for [Computer Vision video summarization].

Really nice product release btw :-)