It's actually quite difficult since it requires many models for each item based on zoom scale progression. Very doable but often skipped, and some are potentially in 5+ zoom levels.
Also hard to reproduce: the selection of objects, the art style (decidedly Japanese) the ambiance (king/prince/planets, the pool table when you lose), music and sounds, and a lot of the little things that make it delightful like things running away, the selection of items - the fact that parts of the world that are not objects become items to pick up at large scales.
Zoom scales sounds like a great job for AI!
Isn't model LoD a... dare I say "solved" problem where basically every video game that could use it, uses it and the generation is automated?