While video generation models produce visually convincing movements, visual quality does not correlate with practical executability by robots — an evaluation criterion overlooked by standard metrics.