SIGNS OF LIFE, page-449

Linius can’t create the moments or distinguish them. It works on metadata provided by the likes of IMG, NRL etc etc and uses these tags to create the highlights. These metatags are created manually (NRL has a room full of students tagging moments on the videos) or will use AI/ML from AWS or specialist S/W providers that can recognise key moments and automatically tag them.

This is the expensive part of the process. You need to pass the whole video file through the recognition to get a file which is tagged. Look at AWS rekognition or Catapult if you want to understand more. It’s also why there some much historical commentary about what LNU actually delivers in the process, which is ultimately an ability to create a new file from these multiple tagged files without having to download the full origin file (again, expensive when most is stored in cloud).