OpenAI has launched Sora 2, its new flagship video and audio generation model, describing it as a major leap in simulating physical reality. The company said the system improves on its predecessor by producing more consistent world dynamics and realistic interactions.
“In Sora 2, if a basketball player misses a shot, it will rebound off the backboard,” OpenAI said, contrasting it with earlier systems that often distorted reality to match prompts. The model also generates synchronized soundscapes and allows users to insert themselves or objects into AI-generated scenes with accurate likeness and voice.
Alongside the release, OpenAI introduced a new iOS application called Sora, which lets users create and remix videos, browse a customizable feed, and appear in “cameos” after verifying their identity. The company said cameos were designed to make the platform feel more interactive and social.
OpenAI added that user wellbeing shaped the app’s design. The feed prioritizes creation over consumption, includes parental controls, and restricts content exposure for teenagers. The firm said it is not optimizing for time spent in the app, a common criticism of rival platforms.
Sora 2’s debut marks a competitive step in the growing field of AI video generation, where rivals are also pushing toward models capable of blending entertainment, communication, and creative production.
Source: OpenAI