Just a curious thought.
In theory the current models may by accident get it right. However since there is no detailed information available about what different rooms are required, their function their size etc. The answer is no, not yet at least.
This. SD and most AI models are good at approximation and mimicry. This is why you occasionally get multiple fingers, heads turned backwards, and double torsos. If you’re building a house it’d probably be good inspiration but some things would be unreasonable like a room with no window or 3 rooms sharing a half bath only.
Well, to be fair, Mike Brady was a hot-shot architect and he had 6 kids sharing one bathroom. Maybe OP would be better off with stable diffusion.
IIRC there are AIs/Algos out there for generating D&D dungeon layouts and so on, so there is no reason why not. The key though is that the layout and the contents is tile based, which significantly reduces the complexity. When rooms can only come in whole numbers of units (2m by 4m, etc) then it doesn’t take a lot of finesse to put them together, and humans have been literally pulling dungeons tiles out of a bag for years to the same effect.
If you are looking at making game content, I bet you that someone has already made a generic building layout model that you can use. If you are looking to do 3D work or anything that needs more finesse you might have problems.
