If I do a building cut up in different polygon (UV) sections with a architectural element on each (window, door, beam, pilasters) overlayed in the texture map, how do a do the LODs?
This method does give a fairly high polycount. Does it demand a new UV and texture on the LODs? With the extra draw calls and texture it does seem expensive.

Draugr

make your LOD proxy, basically the silhouette of the building, and bake the high def model you describe to a single sheet, preferably a tiny portion of a large sheet shared for many LODs.

It requires a new material set and a new model to make LODs. If you were going to just use the same model, it would defeat the purpose as mipmaps are already in the main asset. The LOD needs optimized assets for the tiny area of space on screen it will occupy.

Here is an example of a good LOD image. It is from Mafia2, though I don't know who made it or who to credit:



It's all just baked down from the high to a new low proxy.