OpenAI introduces Sora 2, an AI model for video
OpenAI introduces Sora 2, an AI model that generates video and synchronized audio with greater realism, control and style variety. Access begins with limited invitations and restrictions designed to reduce the non-consensual use of faces and deceptive deepfakes.
OpenAI has introduced Sora 2, an artificial intelligence model that generates videos with synchronized audio from user instructions. The company describes it as an evolution of Sora, with improvements in physics, realism, control and style variety.
The model is designed to make videos look convincing while also following the way the real world works more closely. For example, a scene with a ball, a running person or a falling object should show more coherent movement.
It can also create audio synchronized with the images. That lets you generate a scene with dialogue, ambient sounds or effects without producing them separately.
More control over what it generates
Sora 2 is designed to follow user instructions more accurately. You can ask for a realistic scene, an animation or a sequence with a specific aesthetic, and the model will try to maintain those instructions throughout the video.
The tool will be available at sora.com and through a new standalone iOS app. OpenAI also plans to offer it through its API, although that option will arrive later.
Initial access will be limited and available through invitations. This is not yet a general release for everyone.
The risk of putting real faces in video
Realistic video generation raises issues that do not have the same scale in still images. These include using a person's likeness without their consent and creating deceptive videos that appear to show real events.
OpenAI says it tested Sora 2 with internal risk assessment teams, known as red teamers, to identify problematic uses and design safeguards. The company acknowledges that some risks are still emerging and are not yet fully understood.
That is why the launch begins with several restrictions:
- The use of uploaded images that include a photorealistic person is limited.
- Video uploads are not allowed during this initial phase.
- Content involving minors will be subject to especially strict moderation controls and thresholds.
- Access will expand gradually through limited invitations.
What changes for you
If you get access, Sora 2 can help you create clips from a written idea, prepare a scene for a story or explore visual styles without recording and editing every element from scratch. Synchronized audio also reduces part of the follow-up work.
But the restrictions matter: you will not be able to freely use your own videos or just any image of a real person. Availability, permissions and limits will depend on how the rollout develops.
OpenAI presents Sora 2 as a creative tool and as a step toward models that can simulate the physical world more accurately. What to watch now is whether its safeguards can keep pace with the quality of the content and how it adjusts its rules as the model begins to be used at a larger scale.