MagicTime is a model for generating high-quality change videos from text descriptions. By learning the physics of time-lapse videos, it achieves highly realistic simulations of change processes. The model comprises three main components: MagicAdapter, Dynamic Frames Extraction, and Magic Text-Encoder, which effectively understand the change process from text and generate corresponding videos. The project team has also developed a specialized time-lapse video dataset, ChronoMagic, to support change video generation. Comprehensive experimental results show that MagicTime performs excellently in generating dynamic and realistic change videos, providing a new perspective on building physical change simulators.