With a LOT of the Google magic this can probably be implemented in Google Earth.

Microsoft already bought a company that patented this technology. Sorry, it is going to be in Virtual Earth instead.
I cannot find a link, though.

Here it is!
http://www.geotango.com/
Look down at "3d site building"

good find from an unfront page story!

There are animations available on this too:
http://www.cs.cmu.edu/~dhoiem/projects/popup/index.html

In my humble opinion, a lot more than web 2.0 is, speech & image recognition & mastering technology is the future of the internet...

The only problem now is that where web 2.0 is just web 1.0 + factor 1 complexity, speech & image is web 1.0 X factor 1000 or more complexity. Anyway, I strongly believe that this will hapen and be mature in 5 till 10 years.

A lot of the components needed do exist already at the moment. C.f. speech composition in GPS systems, c.f. voice recognition build in in voice switches, c.f. the -very basic- image recognition system build in in the experimantal toolset of Picasa, etc.

Add some new born AI sauce to that, and it will be a beautifull world. Or is this the last step needed to make machines smarter than men (and even more evil...) ?

Agreed with your points, especially on complexity.

How does this compare to SketchUp?
http://sketchup.google.com/3dwarehouse/search?ityp=1

It seems like the same thing... you can use Earth 4.0 + SketchUp to overlay satellite photos on models to produce photo-realistic models which can then be viewed in Google Earth.

Michael: SketchUp allows you do draw a 3D model yourselves by hand. The other tool (not owned by Google) is making an automatic model out of 2D pictures.

In fact they do the same, but the way the tools become the 3D model is completely different. Not only that, there are other differences. Details made with SketchUp can be a lot better than the 2D3D models. On the other side the bigger picture of the 2D3D models can look a lot more realistic.