If you ignore the implementation, its basically a procedural texturing technique? Those are widely used now.
If you're talking about a real time post effect, it would probably be a bit too slow for a few more years.
If you count SLAM techniques that label camera feeds for AR games, those are very close but I dont think most run at a full framerate.
The paper in question can apparently render at 2K and 30fps though. Or at least that’s what the videos claim.
Even real-time today it should be doable if you create your game with that in mind. You really don't need to generate all the textures, just a compact representation of the level which is to be rendered normally after the fact.
We've had procedurally generated worlds for a long time, but this would take it from roguelike top-down or isometric to immersive fps.
If you mean asset / level generation. Then yes. It is the next step in procedural generation imo.