Introducing Sora 2

Sora 2 arrives as OpenAI’s flagship system for generating video and audio together—plus a new “Cameo” feature that lets people insert a real person (or even a pet/object) into AI-generated scenes after a permissioned setup. The pitch is straightforward: Sora 2 is built to make moving images feel more physically grounded, while the app experience turns those capabilities into a social, identity-aware medium rather than a one-off generation tool.

On the model side, Sora 2 is positioned as a major step up in realism and control. It’s described as more robust at physical interactions than earlier video generators, handling complex dynamics—such as collisions and high-motion stunts like gymnastics routines or backflips on a wakeboard—with greater naturalness. It also improves “steerability,” reducing the need to generate video shot-by-shot and enabling longer, more coherent narratives in a single run.

Audio is the headline addition. Sora 2 is the first Sora model in this lineup that simultaneously generates both video and sound. That includes dialogue across multiple languages, multiple speakers, sound effects, and broader “soundscapes,” aiming to make scenes feel complete rather than visually convincing but acoustically empty.

Cameo is the feature meant to change how people participate. After observing a short clip of a person (Bill, Rohan, Thomas are used as examples), the system can place that individual into any Sora-generated environment based on a prompt. The transcript emphasizes that the capability generalizes beyond humans: observing a clip of a pet or object can still allow injection into prompts. Technically, Cameo is framed as emerging from world simulation models, where the observed subject becomes something the system can treat like a token within the prompt.

The product layer then turns these model abilities into a social feed. The Sora app is presented as a familiar profile-and-follow interface, but with AI-generated content filling the feed—posted by humans, generated by AI. Users can create content via a composer, remix existing posts (including turning a fragrance concept into an ad), and participate in trends by riffing on what others share.

Safety and moderation are treated as central to the identity promise. Cameo requires an explicit permission flow: users record a dynamic audio prompt, pass a liveness check involving head movement, and undergo validation to prevent impersonation. Cameo owners can choose who can use their likeness (only themselves, approved people, mutuals, or everyone) and can delete content they authorized. The app also includes age-appropriate policies (including no infinite scroll by default for under-18 users), nudges away from doom-scrolling toward creation, and labeling/provenance measures such as visible watermarks on exports and traceability techniques including C2PA.

Rollout details close the loop: Sora 2 is launching in the Sora iOS app first, initially in the US and Canada, via an invite-based rollout designed to bring friends together. OpenAI also signals broader access through sora.com (web updates), storyboard-style shot control, and an API planned for the coming weeks—framing Sora 2 as both a consumer social platform and a foundation for creator tools and integrations.