OpenAI has launched Sora 2, the latest version of its video generative AI. The company also launched a Sora social app that lets you share AI-generated videos with your friends.
Sora 2
The original Sora generative AI model was launched in February 2024. OpenAI says that Sora 2 is more physically accurate, realistic and controllable than previous versions. It also features synchronized dialogue and sound effects. Plus, you can create with it via the new Sora app. Currently, the app is only available on iOS, and you can only access it with an invitation.
Improvements
OpenAI says that one of the problems with earlier AI models was that they were “overoptimistic.” With previous generative AI, if a basketball player missed their shot, the ball might just spontaneously teleport to the hoop. Now, with Sora 2, when the basketball player misses their shot, the ball will rebound off the backboard. OpenAI acknowledges that Sora 2 still makes plenty of mistakes. However, the new AI model is better at obeying the laws of physics compared to prior systems. As well as generating video, Sora 2 can also create sophisticated background soundscapes. This can include speech, sound effects, and music, all synchronized to the action.
Insert yourself into an AI video
OpenAI says that you can directly inject elements of the real world into Sora 2-generated videos. After “observing” a video of an OpenAI employee, Sora 2 was then able to insert them into any environment that it generated. The result was an accurate portrayal of the person’s appearance and voice.
OpenAI says that this capability is very general and works for any human, animal or object.
Video samples
OpenAI has shared a number of sample videos generated by Sora 2. The company has showcased a two-minute introductory video with human presenters, all of which was generated by Sora 2. There are also some shorter videos, including one where an ice skater has a cat balanced on their shoulders. Another video shows a man doing a back flip on a paddleboard and looks as if it was shot on a smartphone. Also, there is a video where a dog floats through space to catch a ball in a spaceship. There are a number of animation style videos as well.
Sora app
The new app is simply called Sora and is powered by Sora 2. At the moment, it is only available for iOS. Inside the app, you can create videos to send to your friends. You can also remix each other’s videos or discover new videos in a customizable Sora feed. In addition, via a feature called Cameos, you can drop yourself straight into any Sora scene. All you need to do is record a short video and audio recording in the app once to verify your identity and capture your likeness.
What we think
The Sora 2-generated videos that OpenAI has shared so far are undeniably impressive. Many people would struggle to recognize that the people in them were not real. Sora 2 also seems to be able to generate videos that look cinematic, as well as others that look like they were shot on a smartphone. Additionally, Sora 2 can generate perfectly synchronized soundtracks for the videos. It’s an effective demonstration of what the technology can achieve. However, while OpenAI included a section headed “launching responsibly” on its website, this only addresses safe use of the app and harmful content. There is nothing about the material used to train the Sora 2 generative model. As such, the debate about the ethics of generative AI is set to continue.
Availability
The new Sora app is available to download from the Apple App Store now. However, you will need an invite code to be able to access it at the moment. OpenAI says that it will initially roll out Sora 2 in the U.S. and Canada with the intent to quickly expand to additional countries.
