On Tuesday, OpenAI introduced Sora 2, the next generation of its audio and video creation tool, following last year’s Sora. Alongside the new model, the company unveiled a companion social app, also named Sora, which lets users create videos featuring themselves and friends to share on a feed similar to TikTok’s algorithm-driven stream. Wired had previously reported on OpenAI’s development of this new social platform.
Although we haven’t yet had the chance to try the invite-only app or the Sora 2 model, OpenAI has released some impressive sample videos. Notably, Sora 2 demonstrates improved adherence to physical laws, resulting in more lifelike videos. Publicly shared clips from OpenAI include scenes such as beach volleyball, skateboarding stunts, gymnastics, and people jumping off diving boards, among others.
“Earlier video models tend to be overly optimistic—they might distort objects or reality to fulfill a text prompt,” OpenAI explained in a blog post. “For instance, if a basketball player misses a shot, the ball could magically appear in the hoop. With Sora 2, if a player misses, the ball bounces off the backboard as it would in real life.”
The Sora app features an “upload yourself” option called “cameos,” enabling users to insert themselves into any video generated by Sora. To use their own image in these videos, users must provide a one-time video and audio recording to confirm their identity and capture their likeness.
This tool also lets users share their “cameos” with friends, granting others permission to include their likeness in group videos or other generated content.
“We believe that a social app centered around the ‘cameos’ feature offers the most engaging way to experience Sora 2’s capabilities,” the company stated.
The Sora app for iOS is now available for download, initially launching in the U.S. and Canada. OpenAI has plans to expand to additional countries soon. While the social platform is currently accessible by invitation only, ChatGPT Pro subscribers can access the Sora 2 Pro model without needing an invite.
After creating videos, users can post them to a feed within the Sora app, which appears to function similarly to TikTok, Instagram Reels, or other short-form video platforms. Interestingly, Meta just announced last week the addition of a “Vibes” video feed to its Meta AI app, which is essentially a stream of casual content.
To personalize its recommendations, OpenAI will take into account a user’s Sora activity, geographic location (determined by IP address), previous post interactions, and ChatGPT chat history, though this can be disabled. The Sora app also includes parental controls via ChatGPT, allowing parents to set limits on endless scrolling, disable personalized recommendations, and control who can send messages to their child. However, the effectiveness of these controls depends on the parent’s technical skills.
At launch, the Sora app will be free to use, with OpenAI stating this is to let users explore its features without barriers. The only planned monetization at launch is to charge for generating additional videos during periods of high demand.
Launching a social platform means OpenAI must implement strong user safety protocols, an area where it has faced challenges with ChatGPT. Although users can withdraw permission to use their likeness at any time, this access can still be misused. Even if someone trusts a friend with their AI likeness, that friend could create misleading or harmful content. The issue of non-consensual AI-generated videos remains a significant concern, with limited legal frameworks addressing platform accountability.
