OpenAI Unveils Sora 2: Advanced Video Generation Model with Enhanced Realism and Safety Features

OpenAI has launched Sora 2, its state-of-the-art video and audio generation model. Building on the initial Sora model, Sora 2 introduces improved capabilities that have been difficult for previous video models, including more accurate physics, enhanced realism, synchronized audio, greater user control, and an expanded range of stylistic options. The model is designed to follow user instructions with high fidelity, enabling the creation of imaginative videos grounded in real-world dynamics. OpenAI envisions Sora 2 as expanding the toolkit for storytelling and creative expression, representing a step towards models that can simulate the complexities of the physical world more accurately. Sora 2 will be accessible via sora.com, a new standalone iOS Sora app, and through the OpenAI API in the future. Recognizing the potential risks associated with Sora 2's advanced capabilities, such as nonconsensual use of likeness or misleading generations, OpenAI has implemented various safety measures. These include working with internal red teams to identify new challenges and inform corresponding mitigations. OpenAI is taking an iterative approach to safety, focusing on areas where context is especially important or where risks are still emerging. The initial deployment of Sora 2 will involve limited invitations, restrictions on image uploads featuring photorealistic people and all video uploads, and stringent safeguards and moderation thresholds on content involving minors. OpenAI will continue to learn from user interactions with Sora 2 and refine the system to balance safety and creative potential. A system card detailing the model's capabilities, potential risks, and safety measures has been developed for the deployment of Sora 2.