OpenAI has launched Sora as its new flagship model for video and audio generation. It is said to be a major step forward from the original Sora, which was launched in February last year. The company says that the new system not only improves realism but also introduces new ways for users to create and share content through a dedicated iOS app.
A leap in video simulation
According to the company, Sora 2 is the GPT 3.5 moment for video. Unlike earlier video systems that bent reality to fulfil prompts, Sora 2 is designed to respect physical rules. This means that if a basketball player misses a shot, the ball will bounce off the rim instead of teleporting to the hoop. Apart from that, the model is also said to simulate complex stunts, such as Olympic gymnastics routines, paddleboard backflips, or even a triple axel with a cat hanging on.
More control, more realism
Sora 2 also improves controllability, following complex prompts across multiple shots while keeping details consistent. It supports various video styles, including realistic, cinematic, and anime. It can also generate high-fidelity background audio, speech, and sound effects.
Apart from that, users can even inject real-world elements into generated content. With just a short video, the model can insert someone into any scene, replicating their appearance and voice. OpenAI refers to this as the “cameos” feature.
The new Sora app
Along with the new model, OpenAI is launching Sora, an invite-only iOS app built around Sora 2. The app allows users to create videos, remix each other’s work, and use Cameos to appear in different scenes. Testers inside OpenAI say the feature has already helped colleagues connect in new ways.
OpenAI stresses that the app is designed for creation rather than endless scrolling. The feed is shaped by natural language-driven recommender algorithms that prioritize inspiration over watch time. For teens, the app features strict content limits, parental controls through ChatGPT, and safeguards against the misuse of likeness.
Responsible rollout
The company states that it is taking steps to address concerns regarding safety, well-being, and content misuse. Users control their likeness at every step and can revoke cameo permissions or remove videos at any time. OpenAI is also scaling up moderation teams to handle cases of bullying and harmful content.
For now, Sora 2 will be free to use with generous limits, though usage may eventually require payments tied to compute demand. ChatGPT Pro subscribers will get access to a higher-quality “Sora 2 Pro” model. An API release is also planned.
Availability
The Sora app is available today on iOS in the U.S. and Canada, with plans to expand to other regions soon. Users can request access via the app or sora.com. Existing Sora 1 Turbo content remains accessible in user libraries.
