OpenAI today released Sora 2, a major upgrade to its video-and-audio generation model, alongside a dedicated Sora iOS app that functions as a TikTok-style social network. The app is available to download now and rolls out first in the U.S. and Canada, with the social feed invite-only at launch. It offers an algorithmic feed of user-generated clips and a ‘cameos’ capability that lets people upload a one-time video-and-audio recording to verify their identity and insert their own likeness — or a friend’s, with permission — into any generated scene.

Sora 2 marks a significant leap in physical realism, the company claims. In a blog post, OpenAI said prior video models were ‘overoptimistic,’ morphing objects and deforming reality to satisfy a text prompt. The new model, it says, ensures that if a basketball player misses a shot, the ball rebounds off the backboard rather than teleporting into the hoop. Sora 2 also generates synchronized dialogue and sound effects alongside video.

The Sora app is free at launch, with monetization planned only to charge for extra generations during periods of high demand. OpenAI said the app ships with parental controls managed through ChatGPT, including the ability to override infinite scroll limits and manage who can direct message a teen. The company acknowledged the potential for misuse, particularly the cameo feature: even someone a user trusts with access to their likeness, it noted, could still generate deceptive content. The launch comes days after Meta added a similar video feed called ‘Vibes’ to its AI app.