Training a video generation model from scratch on limited hardware such as a laptop has an extremely low return on investment, and a more pragmatic approach is to use transfer learning and fine-tuning.