OpenAI Introduces Sora 2 Video Generation AI Model, Launches New TikTok-Like Social App

OpenAI unveiled the Sora 2 video generation model on Tuesday, introducing several new capabilities and improvements. The artificial intelligence (AI) model arrives nearly a year after the first Sora model was released. Access to the new AI video model will be available for free via a new TikTok-style social media app. The iOS-only app's access can only be obtained via an invite. The app allows users to generate AI videos of themselves and share them with friends. Users can also generate similar videos of others via a “Cameo” feature.

Sora 2 AI Model and New Sora for iOS App Released

In a blog post, the San Francisco-based AI firm announced and detailed the new AI model and the app. The biggest improvement comes in terms of native audio generation, bringing it on par with Google's Veo 3. Calling Sora 2 the “GPT‑3.5 moment for video,” OpenAI said that it can now handle complex movements while maintaining realistic physics in videos.

In a demo video, a gymnast was shown performing flips on a balance beam with close-to-reality movements, highlighting the capabilities of the model. OpenAI also claimed that Sora 2 does not morph objects or deform reality to execute a prompt. This would mean fewer hallucinations in the output videos.

Controllability of the output has also been improved, and now Sora 2 can follow instructions spanning multiple shots while keeping the world state consistent, the company said. Notably, the video generation model is said to perform the best when it comes to realistic, cinematic, and anime styles.

Currently, the Sora 2 model is available via the Sora for iOS app. It is an invite-only app, which is available to individuals in Canada and the US. The new app resembles TikTok, Instagram Reels, and YouTube Shorts in terms of interface. The only difference is that all the videos on this app are generated by Sora 2.

Notably, Sora 2 video generation is currently free for all users with an invite. Also, after a user has received the invite to the app, they can also visit the Sora website to generate videos.

Users can upload an input video of themselves, and using its high character consistency, the AI model can place the character in any other setting. Users can then post these videos, and their followers can watch, like, comment, and share them. They can also “Cameo” them. While scrolling through the videos, a user can tap on the Cameo button to take the character of that video and generate a different video using them and a text prompt. Essentially, it lets users create deepfakes of their friends.

“We first started playing with this “upload yourself” feature several months ago on the Sora team, and we all had a blast with it. It kind of felt like a natural evolution of communication—from text messages to emojis to voice notes to this,” the company said.

Notably, before a user generates a video of themselves using the app, they will have to verify their identity. So, a person cannot use a random video of a celebrity and then generate videos of them. However, the Cameo feature does take away some of the safeguards.

Interestingly, OpenAI is not the only one exploring the vertical scrolling AI-generated videos app idea. Last week, Meta launched the Vibe feed on the Meta AI app, allowing users to watch short videos generated using in-house AI models.