I've been curious about what the bandwidth would be like to image recognize my facial expression, encode it and then just send the event stream to a virtual avatar.