I see one potential downside of this, as a Netflix subscriber I do find the frequent shuffle and changes of cover art confusing sometimes as it makes you feel there is a new film coming out but in fact you've already watched it.
You can setup AVA style video processing pipeline using my project <a href="https://www.deepvideoanalytics.com/" rel="nofollow">https://www.deepvideoanalytics.com/</a><p>It comes built in with face detection/recognition, object detection, OCR, Open Images tagger etc.
Maybe they're not telling about some of the detail (and they've clearly given it plenty of thought!) but it seems wasteful to fully process all frames when in several cases many of them will never be selected. eg if it fails due to poor Visual Metadata (such as strong motion blur or very low brightness) you'd want to abandon all frames like that before running the other processing.