It looks human-janky instead of ai-janky. Like the fingers are “person just sort of going ‘there are fingers here’” instead of the mix of too detailed and yet still fundamentally wrong that AI does. It’s also composited with what are clearly photos of objects. On top of that the characters are low-detail, looking at objects in the scene instead of at the viewer/someplace random, and they have actual expressions displayed with abstract details instead of being either exaggerated or blank.