Video on YT: https://www.youtube.com/watch?v=kPMHcanq0xM
Company: http://en.wikipedia.org/wiki/Weta_Workshop , http://en.wikipedia.org/wiki/Weta_Digital
It's called augmented reality and more basic implementations are already possible for some time: http://en.wikipedia.org/wiki/Augmented_reality
@Downvoter: what's wrong?
The video credits list Weta prominently - that's alone is a significant signal. The guns are props from Weta too: https://www.wetanz.com/rayguns/
CEO Rony Abovitz, who was supposed to go onstage at the TED conference in Vancouver on Wednesday, has canceled his talk. Why would a CEO cancel such a high profile talk and release that video afterwards? More info: http://recode.net/2015/03/16/magic-leap-googles-high-profile... , http://recode.net/2015/03/19/the-first-video-of-magic-leaps-... .
I tested two state of the art augmented reality glasses (university research prototypes) last year. The glass was connected (via cable) to a workstation PC with several high end NVidia graphic cards. One also used Xbox Kinect to track the environment and the arms. Everything shown in the video is already possible, but not with that perfect visual quality and without any latency.
It's worth noting Peter Jackson has made an entire documentary on a forgotten genius of the New Zealand film industry that turned out to be a (good-natured) hoax. I'm perfectly prepared to believe the guy responsible for Forgotten Silver would have a team around him who might do the same for VR.
EDIT: here's a particularly blatant example, around 0:17: 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.
Edit: Actually, no, I don't think it's right. Modern chroma keying could probably handle this.
As far as the grunting, how is it any different than having a Nerf gunfight or pingpong or whatever people are into at the office?
Especially when the actor scrolls through his mail, his hand is a very low resolution and the mails are a very high resolution.
I suspect this is a major signal of it being fake.