48 tasks, 4 models, one pattern: everyone's better at looking right than moving right. Visual similarity beats motion consistency 42-46 out of 48 times across the board — the next frontier isn't sharper pixels, it's coherent motion.