A new AI system developed by researchers at Stanford University demonstrates the ability to generate realistic video from text descriptions. The model, named VideoGen, uses a novel diffusion architecture to create short, coherent clips based on prompts like “a cat playing with a ball of yarn.” While the technology shows significant progress in temporal coherence and visual quality compared to previous systems, the researchers acknowledge limitations in resolution and the complexity of scenes it can handle. The team emphasizes the potential applications in film pre-visualization and educational content, while also highlighting the need for robust detection methods to mitigate risks of misuse, such as generating deepfakes. The full research paper is available for review at the conference proceedings link.