OpenAI has unveiled Sora 2, its next-generation video and audio generation model. Open AI describes Sora 2 as a leap comparable to the jump from GPT-1 to GPT-3.5 in natural language AI.
The company says the model demonstrates advanced world simulation abilities. It obeys the laws of physics more consistently than earlier systems. Additionally, it introduces stronger controllability for complex, multi-shot scenes.
About Sora 2
Unlike previous video generators that often morphed objects or “cheated” to fulfil prompts, Sora 2 is capable of modelling failure realistically. For instance, it can show a basketball bouncing off a backboard rather than teleporting into a hoop. It also excels across styles, from cinematic realism to anime. Furthermore, it integrates soundscapes, speech, and effects with high fidelity. Users can even insert themselves into Sora-generated videos through a feature called cameos. This feature builds a likeness and voice profile from a short one-time recording.
To showcase the technology, OpenAI is launching Sora, a new invite-based iOS social app powered by Sora 2. The app allows users to create and remix videos, discover content through a customizable feed, and star in friends’ creations through cameos. OpenAI stresses that the app is designed for creativity and community rather than passive consumption. It includes built-in wellbeing checks, parental controls, and safety features around likeness, consent, and harmful content.
Initially rolling out in the U.S. and Canada, Sora will expand to more countries soon. The app and sora.com will provide free access with “generous limits,” while ChatGPT Pro users can try an experimental higher-quality version of Sora 2 Pro. OpenAI says it eventually plans to make the model available through its API.
