Researchers have developed a new AI model capable of generating highly realistic and coherent video from simple text descriptions. The system, named ‘CineGen’, uses a novel diffusion-based architecture that builds scenes frame-by-frame with a strong understanding of physical dynamics and object permanence. Early demonstrations show the model creating short clips of complex scenes, such as animals in motion or changing weather patterns, with significantly fewer visual artifacts than previous methods. The team highlights potential applications in film pre-visualization, game development, and educational content, but also acknowledges serious concerns regarding the creation of deepfakes and misinformation. They are releasing the model with usage restrictions and are advocating for industry-wide standards on synthetic media. For the full details and video examples, read the complete article at https://technologyreview.com/2024/05/15/cinegen-ai-video-generation.