A grainy video has just become available from the Adobe Max conference in Milan, Italy last month of a sneak peek at a new experiment called Infinite Images.

The project is so cool we couldn't help but post about it, even though very little about it is known yet.

Infinite Images takes any collection of tagged images, not necessarily from the same location at all, and stitches them together in 3D by analyzing their composition and the semantics of their tags ("sky" is above "ground," for example). Microsoft's PhotoSynth is cool, but this is much cooler.

The results can be an imaginary panorama made up of a logically combined set of images from different photos, or a zoomable 3D world reminiscent of game rendering engine output but built out of photos uploaded to the system.

Designate one image as the front of a zoomable space and another the back and Infinite Images can assemble a whole continuous landscape between them.

Check out this video demonstration below. The action starts at about 1:00 in. We're not sure what kinds of use cases this might lead to, but it's very exciting. It's just one more example of the kinds of exciting things that can be done with semantic data; who knows what people will think up next?