OpenAI’s Sora 2 is a groundbreaking AI video and audio generation model that introduces lifelike physics, ‘cameo’ social creation, and responsible design principles—pushing AI closer than ever to full world simulation. (Source: Image by RR)

Model Can Accurately Simulate Physical Motion, Failure, and Multi-Shot Continuity

OpenAI has released Sora 2, a major upgrade to its flagship video and audio generation model, marking what the company calls the “GPT-3.5 moment for video.” Building on the original Sora’s early success in simulating motion and object permanence, Sora 2 introduces far more realistic world modeling—accurately portraying physics, mistakes, and continuity across scenes. The model can now handle highly complex actions such as Olympic-level gymnastics routines, realistic basketball rebounds, and dynamic multi-shot sequences while maintaining consistent environments and lighting. Sora 2 also integrates lifelike soundscapes, dialogue, and ambient audio with video in a seamless generation process, pushing OpenAI closer to a full world-simulation engine capable of understanding physical reality.

A standout new capability, as noted in openai.com, allows users to inject real-world likenesses into AI-generated videos. With just a short video and voice sample, Sora 2 can place a user—or anyone—into scenes with realistic speech, movement, and emotional nuance. This “upload yourself” feature forms the basis of OpenAI’s new iOS app, “Sora,” which launches alongside the model. The app is designed for users to create, remix, and collaborate through “cameos,” where friends can drop themselves into shared scenes. OpenAI says the app aims to feel more like “a creative evolution of communication” than another social feed, prioritizing artistic play over passive scrolling.

To address safety and wellbeing concerns, OpenAI has built new AI-curated feed controls powered by natural language prompts, letting users guide what appears on their feeds. The company stresses that Sora’s recommender algorithms are not optimized for time spent but for creative engagement. The app includes parental controls, wellbeing check-ins, and teen-specific limits, alongside full control over one’s likeness. Users can revoke cameo access or delete any video containing them. A detailed Sora 2 Safety framework outlines content moderation, consent protocols, and provenance tracking for generated media.

Sora 2 is now rolling out in the U.S. and Canada, available through the Sora iOS app and sora.com, with a free tier and generous usage limits. ChatGPT Pro users will gain access to Sora 2 Pro, a higher-quality experimental version, and an API release is forthcoming. OpenAI frames Sora 2 not just as an entertainment platform but as a milestone toward general-purpose world simulation—a technology it believes will underpin the next generation of AI agents and robotics. “We think Sora is going to bring a lot of joy, creativity, and connection to the world,” the team wrote, calling it the start of a “new era of co-creative experiences.”

read more at openai.com