Using time-lapse videos as training data, computer scientists have developed text-to-video generators that simulate the physical world more accurately.