Sora 2 is OpenAI's second-generation large-scale video generation model, officially released in October 2025 as a successor to the original Sora research preview. Positioned as a multimodal "world simulator," the model produces high-fidelity video content from text prompts and images with significantly improved physical realism, temporal coherence, and resolution compared to earlier iterations.
A defining feature of Sora 2 is the native integration of synchronized audio generation. This capability allows the model to generate dialogue, ambient soundscapes, and music that are automatically aligned with visual actions, including accurate lip-syncing for characters. It also introduced the Cameos feature, which enables users to insert consistent personas or their own verified likeness into generated scenes, facilitating personalized storytelling.
From a technical standpoint, Sora 2 demonstrates an advanced understanding of real-world physics, accurately simulating complex interactions such as object rebounds, fluid dynamics, and lighting effects. The model supports various aspect ratios and delivers outputs up to 1080p resolution. While the original Sora was primarily a research project, Sora 2 was launched as a consumer-facing product integrated into a dedicated social application and the sora.com web platform.