Sora: OpenAI shows minute-long AI video from a prompt
In one sentence OpenAI announces Sora, a text-to-video model that generates clips up to one minute long while keeping visual quality and prompt adherence. Initial access only for red teamers and a group of visual artists, designers and filmmakers.
OpenAI publishes a gallery of striking videos, including a woman walking down a Tokyo street, generated from a single text prompt and up to one minute long.
Visual quality is clearly above the text-to-video tools available at the time. The model seems to "understand" how people, objects and the camera move, but OpenAI also lists its limits: for example a person can bite a cookie that then stays intact.
It is not public: at first only red teamers, looking for risks and misuse, and some artists, designers and filmmakers giving feedback can use it.
Companies
OpenAI
Tools
Sora
Tags
Sources