OpenAI has released Sora, breaking open a new evolution of the AI model: video generation. How does it create its intensely realistic videos? What are the risks and flaws in this monumental step in AI technology?
OpenAI has unveiled Sora, a groundbreaking AI video generator capable of transforming text prompts and still images into high-quality videos nearly indistinguishable from reality. Building on the foundation of text-to-image generative models, Sora organizes three-dimensional “patches,” adding time as a dimension, to generate videos frame by frame. To manage its massive computational needs, the model uses dimensionality reduction and recapturing techniques, ensuring efficient processing without sacrificing resolution. Users can access Sora through premium plans, but it still has limitations, such as struggles with continuity and cause-effect relationships.