Remembering off-screen objects, generating spatially consistent features, modeling physical interactions and lights, understanding what "up the stairs" means, a

Remembering off-screen objects, generating spatially consistent features, modeling physical interactions and lights, understanding what "up the stairs" means, all seem to warrant talking about a world model, because that's exactly what's required to do these things compared to simply hallucinating video sequences.