OpenAI Unveils Sora 2 – AI Video Generator With Realistic Physics And Logic
Credit: OpenAI
American company OpenAI, known for developing ChatGPT, has unveiled the second generation of its Sora 2 video-creation neural network and an iOS app of the same name. The announcement was published on the company’s official blog.
While earlier video generation models often created a believable “picture” but struggled with basic motion logic—for example, they could “teleport” a basketball into the hoop after a miss—Sora 2 models the behavior of objects. A miss means the ball will bounce off the backboard. A figure skater attempting a triple axel might make a mistake and fall. The system has learned to simulate not only success but also failure—a key requirement for creating realistic simulations and advanced robots. The developers promise that there will no longer be strange object deformations or violations of scene logic for the sake of adhering to the prompt.
As a universal video and audio generation system, Sora 2 is capable of creating complex background soundscapes, speech, and sound effects with a high degree of realism. A short video recording is all it takes: the model will accurately reproduce the user’s appearance, facial expressions, and even voice, seamlessly integrating them into any scene. This capability is universal and works with any person, animal, or object, according to the OpenAI press release.
Currently, access to the products is by invitation only for residents of the US and Canada.
Android device users will be able to work with the neural network via a web interface after receiving access. The service is initially free, but the company has not yet disclosed any potential limitations.
