OpenAI is turning its video generation work into a consumer social product. On Tuesday, the company announced Sora 2, an audio and video generator that follows last year's Sora, and introduced a connected iOS app called Sora.
The app is designed around AI-generated clips that can be shared in a TikTok-style algorithmic feed. Its most distinctive feature is called cameos, which lets users place themselves, and potentially their friends, inside Sora-generated scenes.
What Sora 2 changes
OpenAI is positioning Sora 2 as a more realistic successor to Sora. The company has shared examples showing beach volleyball, skateboard tricks, gymnastics routines, and cannonball jumps from a diving board, among other scenes.
The key improvement described by OpenAI is physical consistency. The company says earlier video models could force a prompt to succeed by changing objects or bending reality, even when the result made little sense.
"Prior video models are overoptimistic — they will morph objects and deform reality to successfully execute upon a text prompt," OpenAI wrote in a blog post.
OpenAI gave a basketball example to explain the difference. In an older model, a missed shot might still end with the ball appearing in the hoop. In Sora 2, the company says a missed shot should instead rebound off the backboard.
That matters because short video depends heavily on motion. A clip can look convincing for a moment and still fail if bodies, balls, boards, water, or other moving objects behave in ways viewers instantly recognize as wrong. OpenAI's examples suggest the company wants Sora 2 to compete not only on visual quality, but on whether generated action feels coherent from one frame to the next.
The Sora app brings AI video into a feed
The Sora app is not just a place to generate clips. It is also a social platform where generated videos can be posted into an algorithmic feed similar in broad shape to TikTok, Instagram Reels, and other short-form video products.
The app is invite-only for now. The Sora iOS app is available to download and will initially roll out in the U.S. and Canada. OpenAI says it hopes to expand quickly to other countries.
ChatGPT Pro users should be able to try the Sora 2 Pro model without an invite. That creates two related but separate paths: access to a more advanced model for some ChatGPT Pro users, and access to the Sora social app through the invite-only rollout.
The launch also lands in a broader moment where AI companies are experimenting with feeds built around generated media. Meta announced just last week that it added a video feed called "Vibes" to its Meta AI app.
Cameos make identity the center of the product
Sora's most important product idea is cameos. The feature lets users upload themselves once, then drop their likeness into Sora-generated scenes.
To use their own likeness, users must upload a one-time video-and-audio recording. OpenAI says this verifies identity and captures the user's appearance.
Cameos can also be shared with friends. That means a user can give another person permission to include their likeness in generated videos, including videos with multiple people together.
"We think a social app built around this ‘cameos’ feature is the best way to experience the magic of Sora 2," the company wrote.
That design makes consent central to how the app works. A generated video platform built around likenesses is different from a standard camera app or feed because a person's image can appear in scenes they did not film. The source article notes that users can revoke access to their likeness at any time, but it also highlights how easily this kind of access can be abused.
Personalization, controls, and monetization
OpenAI will use several signals to shape Sora's algorithmic recommendations. Those include a user's Sora activity, their location through IP address, past post engagement, and ChatGPT conversation history. The ChatGPT conversation history signal can be turned off.
The app also includes parental controls through ChatGPT. Parents can override infinite scroll limits, turn off algorithmic personalization, and manage who can direct message their child.
Those tools may help some families set boundaries, but their practical effect depends on whether parents know how to use them. The source article notes that these features are only as powerful as the parent's technical know-how.
Sora will be free at launch. OpenAI says this is "so people can freely explore its capabilities." The company's only stated monetization plan at launch is to charge users to generate extra videos during times of high demand.
The safety challenge ahead
A social app based on AI-generated video gives OpenAI a different kind of responsibility than a standalone generation tool. Once videos can be generated, personalized, and distributed through a feed, user safety becomes part of the product's core design.
The main risk described in the source article is misuse of a person's likeness. Even when someone grants access to a friend, that permission could still lead to deceptive content that harms the person represented in the video.
Non-consensual videos remain a persistent problem with AI-generated video. The source article also notes that few laws explicitly govern platform responsibility in this area.
That leaves OpenAI launching Sora with a difficult balance to strike. The company is offering a free, invite-only social app built around powerful video generation and personal cameos. At the same time, the features that make the app compelling also create new pressure around consent, trust, direct messaging, recommendation systems, and likeness control.
Sora 2 may improve how generated action behaves on screen. The larger test is whether the Sora app can turn that capability into a social product without letting identity-based misuse become part of the feed.