OpenAI has officially unveiled Sora 2, the latest version of its AI video and audio generator, marking a major leap forward in realism, control, and physical accuracy. According to the company’s announcement, Sora 2 builds upon the original model’s capabilities by introducing more sophisticated simulations that respect real-world physics—an area where earlier models often struggled.
The new model can now produce highly dynamic scenes involving complex movements such as gymnastics routines, paddleboard backflips, and triple axel figure skating jumps, all while generating synchronized sound, dialogue, and ambient effects. Another major addition is the ability to seamlessly insert real people or objects from video clips into AI-generated scenes, recreating both their likeness and voice with impressive precision.
Alongside Sora 2, OpenAI is launching a dedicated Sora app for iOS, designed to make AI video creation more social and collaborative. Users can generate and remix AI videos, share their work, and use a “cameo” feature to insert themselves—or friends—into other creators’ clips. If your likeness appears in a remixed video, you’ll be listed as a co-creator and can request its removal, ensuring tighter creative control.
OpenAI says the app experience will emphasize creation over consumption, providing features that promote responsible use and reduce addictive scrolling. Initially, the Sora app will launch in the US and Canada, while Sora 2 will be available on the web through invite-only access.
