Lighter version: stick figures of the person and stick of the gifs (pre process the stick figures by analyzing your videos on the server, not real time, only the user in real-time)
Initially, I wanted to go way further and add 3D avatars dancing like the connected users. This would use the webcam + bodypix to map the user dance with its 3D avatar. However, all of this requires just too much computation on client side to have something useable. Anyway, any suggestions on that lighter version?