But for the moment, I’m guessing there must be a considerable amount of post-processing, touch-up on the textures, and so on. As well as animation, I imagine it won’t do any of that fancy lighting they show, for example.

I came in here to say that I want this game. In my mouf. Right here. :U ←

I just hope that the gameplay and other materials are decent enough as well, because innovative asset-creation techniques aren’t enough to make a game great. But if they make a good game underneath the lovely graphics, I’m in. I’ll buy copies for friends.

Theoretically it should be possible to have someone basically walk through an area collecting the data and have the area then recreated in the gamespace automatically, with the caveat that complex shapes may not be rendered particularly well.  Simple polygons and curves should be easy enough to do automatically though.

The big downside is textures, and particularly texture memory.  This setup would create unique textures for every single wall in a house for instance, which is inefficient and won’t work with today’s graphics cards.

That said, there are duplicate recognition algorithms available (people use them to cull duplicates from large collections of images, even with the duplicate is in a different resolution and quality) that should be able to handle this problem somewhat. 

It would be an incredibly complex tool, but if done right it could save companies millions of dollars in development costs, especially for spunkgargleweewee games that are focused on realism in realistic settings. 

If you want to explore similar tools for free or cheap, check out things like Autodesk’s 123D Catch, or Agisoft PhotoScan – hell, even Microsoft’s Photosynth uses the same back-end math.

I’m not sure what you’re proposing would save a game studio any money, though.

Video games could be made like movies, with location scouts and physical design.  Maybe with some scenes made the old way when a particularly fantastical element is needed.