dumb question: could we, in the future, use some kind of gen ai to generate a videogame map (i'm thinking quake 3 arena / openarena) of buildings like these ?
I think the biggest challenge is not any of the technical or legal problems already mentioned, but that none of these buildings are laid out with the primary objective of being fun to run around shooting people in. So once the novelty wears off, I expect the actual gameplay experience will be rather clunky, especially with competitive gamers.
With players in control any jank will be quite obvious, the field did accelerate thanks to neural nets but there seems to have been a lot of focus on NeRFs and GS (This interactive demo seems to use GS) and classic triangle-geometry (especially lower polygon counts) hasn't gotten as much love recently as the impressive GS demos has taken over.
But the success of GS and speeding up should rekindle some interest and let us use some of the advances in making "production ready" methods.
Whether you can make reproductions of buildings and public interiors is known as "freedom of panorama". Wikimedia Commons has a comprehensive list by country [1].
I wouldn't care about reproducing existing building as long as the AI can generate credible ones in the same style, then place them on dynamic worlds created by prompting the AI. Having intelligent AI NPCs as long as AI generated scenery would be a killer feature in any game. I'm talking about off line disconnected single player games; cloud ones with these features could be already here, but I want to be in control, and marketing rules are against that: who would buy the new shiny V2.0 with the new worlds and characters if 1.0 could create them just by asking it to?
As someone who consumed tons of scifi novels and books as a kid and wants to be immersed in big worlds, enjoying great stories also in games (absolutely loved the Mass Effect saga), I already know what's going to happen when we'll be able to feed Philip K. Dick or Asimov, Sturgeon, Bova, Silverberg, etc. books to an AI and have it create worlds, environments, stories and characters straight out of the book descriptions. Literally drooling over it.
It's like that kid that got expelled for creating a map of his school in Counter-Strike [1], due to fears of security threats. Not that I blame them, I could see people planning a robbery in Minecraft.
To further sustain this point: I heard that in the past, someone recreated some parts of Politecnico di Milano (a famous technical university in Italy) as a map of some open source first person shooter. Unfortunately I don't remember which shooter it was.
One such pipeline that already works today is photogrammetry of real place -> voxel data using VoxelPlugin. You can then leave it as a Voxel or bake it to a static mesh.
(not just the basilica di san pietro)