The interface widgets appear to be occluded by the user's hand, with no visible artifacts even when the hand is moving quickly. That means the system has to be

The interface widgets appear to be occluded by the user's hand, with no visible artifacts even when the hand is moving quickly. That means the system has to be tracking the hand and somehow separating it from the background, with latency of much less than a single frame. EDIT: here's a particularly blatant example, around 0:17: http://i.imgur.com/8mSv8oC.jpg http://i.imgur.com/8mSv8oC.jpg The user is swiping from right to left, causing their hand to be highly motion-blurred. But the rendered UI elements smoothly fade in behind the hand. There are foreground segmentation algorithms that can do this kind of thing, assuming you have some information about which regions of the image are foreground/background (say, from a depth sensor). But those algorithms typically take on the order of seconds or minutes to process a single frame. For this video to be real, the total latency -- sensing, matte generation, and rendering -- would have to be in the single-digit milliseconds. It's barely on the edge of plausibility, but I'm very very skeptical.