OpenAI Unveils Sora, an AI That Can Generate Photorealistic Video from Text
-
OpenAI unveiled Sora, an AI system that can generate photorealistic video from text prompts by learning about 3D geometry and consistency from data.
-
Sora is an evolution of diffusion transformer models, trained by adding noise to images/video and learning to remove it to generate new content.
-
Sora shows emergent capabilities like simulating people, animals, and environments, suggesting it understands aspects of the 3D physical world.
-
Limitations exist like not fully capturing cause-and-effect, but Sora hints at future AI-generated video being indistinguishable from real.
-
OpenAI is carefully assessing risks, slowly rolling out access to “red teamers” given potential for misuse.