To create GIFs, you will have to move in front of the webcam and Move Mirror will match your moves with a database of 80,000 photos, in real-time. To match your moves the pictures comprise all kinds of human activity such as acting and martial arts, among others. Because of which your GIF will look like a clip of different people, performing your dance moves. A machine learning-based computer vision model called PoseNet is used by Move Mirror which detects human figures in images and videos by identifying where key body joints are. It is powered by Tensorflow. On your device machine learning models like PoseNet are run by Js library. This means all processes happen locally on browser and nothing is stored or sent to a server. As a part of Google's AI Experiments initiative Move Mirror aims to simplify machine learning for people by helping them explore and understand the technology via simple experiments.