The field of video generation is moving towards incorporating physical laws and identity preservation into generated videos. Researchers are exploring the use of reinforcement learning and preference optimization to improve the physics-awareness and realism of generated videos. This includes developing methods to capture physical knowledge and inject it into video generation models, as well as optimizing video generation to preserve human identity and consistency. Notable papers in this area include PhysMaster, which proposes a reinforcement learning approach to improve physics-awareness in video generation, and Identity-Preserving Reward-guided Optimization, which introduces a novel video diffusion framework to enhance identity preservation. Identity-GRPO is also noteworthy for its human feedback-driven optimization pipeline for refining multi-human identity-preserving video generation, and RealDPO for its novel alignment paradigm that leverages real-world data for preference learning.