OpenAI just pulled back the curtain on Sora 2, its new flagship model for AI-generated video and audio and it’s a serious upgrade. The company claims this latest version brings real-world physics, precise sound, and user-controlled cameos into the mix. Alongside the model, OpenAI is also launching a dedicated Sora app that turns AI video generation into a social experience.
Sora 2 raises the bar with real-world physics
Sora 2 doesn’t just look better, it behaves better. According to OpenAI, this model moves beyond prompt satisfaction and focuses on plausible world simulation. Earlier versions would force scenes to match prompts, even if that meant objects were teleported or ignored natural rules. Sora 2 now models realistic outcomes, like:
- Missed basketball shots that bounce naturally
- Olympic-level gymnastics with real-time body physics
- Water-based stunts that reflect actual buoyancy and balance
That realism extends to audio, too. Background soundscapes, dialogue, and effects are now fully synced with the on-screen action, giving AI-generated content a more grounded, immersive feel.
Sora 2 cameo feature lets you upload yourself
One standout upgrade is the “cameo” feature. Users can record a short video and voice clip to insert themselves or their pets, friends, or objects into generated scenes. Visual and vocal accuracy is shockingly tight, and there’s a verification layer to maintain consent and safety.
This feature is baked into the new Sora iOS app, which lets users create, remix, and explore AI videos. Unlike traditional platforms, Sora’s feed is creation-driven rather than algorithm-optimized. You see work from people you follow, not just viral bait.
The Sora app focuses on creativity, not clicks
OpenAI insists this app isn’t chasing screen time. It features:
- Privacy-first cameos with full control
- Teen account protections like daily feed limits
- Parental controls via ChatGPT
- Human moderation for safety and anti-abuse enforcement
The Sora app is invite-only for now, with rollout starting across the U.S. and Canada. The Sora 2 model is free at launch, though usage is subject to compute limits. ChatGPT Pro users can already access the Pro-grade version inside ChatGPT itself.
Sora 2 sets the stage for general-purpose simulation
With Sora 2, OpenAI isn’t just chasing viral videos. The company frames it as an early step toward general-purpose world simulator models that don’t just entertain, but understand the physical world. That kind of foundation, they argue, is necessary for future advances in robotics and embodied AI.
For now, though, it’s about letting people show up in AI videos that actually make sense, and that’s a big enough leap already.