OpenAI unveiled an updated version of the Sora 2 AI model

It’s been a year since the Sora model was announced, and now OpenAI has released its first major version called Sora 2. The new model creates more realistic and physically accurate videos, and offers improved control and advanced editing capabilities. Additionally, sound generation has been improved: speech and sound effects are now synchronized with the image.

The company describes the update as a qualitative leap comparable to the transition from GPT-3 to GPT-3.5. Generated clips can follow complex instructions for multiple scenes and maintain a consistent style and plot. A variety of video formats are supported, including realistic, cinematic and anime.

The main innovations of Sora 2 include more natural, distortion-free object physics, improved sound, and expanded editing and control tools. Users will be able to add their own images or videos as “cameos” with the ability to revoke permission at any time. Protective measures are also built in: images of public people can’t be used without their consent.

The launch of Sora 2 began today and is limited to the US and Canada for now. At launch, the service will be free with certain limits. ChatGPT Pro subscribers will have access to a higher quality version of Sora 2 Pro. OpenAI promises to release an API for developers in the near future as well. There will also be a separate app for iOS, available for now through an invitation system.