The rendering isn't the difficult part. All of the techniques they used to actually make her appear on screen are being used in most modern games (bump maps, normal maps, etc) in real time. The difficult part is in recording, storing and interpreting the thousands of tiny facial movements that a gesture is composed of.