OpenAI announced Tuesday the release of Sora 2, its advanced audio and video generative model, alongside a new social application also named Sora. The app is designed to enable users to generate and share short-form videos featuring themselves and their connections within an algorithmic feed, marking the company's entry into consumer social media platforms. Wired previously reported on OpenAI's development of a social platform linked to its generative AI.
Sora 2 is presented as an advancement over its predecessor, demonstrating improved adherence to the laws of physics and enhanced realism in generated content. OpenAI detailed in a blog post that "Prior video models are overoptimistic — they will morph objects and deform reality to successfully execute upon a text prompt." The company added, "For example, if a basketball player misses a shot, the ball may spontaneously teleport to the hoop. In Sora 2, if a basketball player misses a shot, it will rebound off the backboard." Publicly released examples from OpenAI showcase diverse scenarios, including beach volleyball games, skateboarding, and gymnastics routines.
The companion Sora social application introduces a "cameos" feature, allowing users to integrate their likeness into Sora-generated scenes. Utilization of this feature requires a one-time video and audio recording for identity verification and appearance capture. This functionality also permits users to grant friends permission to include their likeness in videos, facilitating the creation of multi-person generated content. OpenAI stated, "We think a social app built around this 'cameos' feature is the best way to experience the magic of Sora 2."
The Sora iOS app is currently available for download, with an initial rollout in the U.S. and Canada, and plans for rapid international expansion. While the social platform operates on an invite-only basis, ChatGPT Pro subscribers are granted access to the Sora 2 Pro model without an invitation. OpenAI plans for the app to be free at launch, with potential monetization through charges for additional video generation during periods of high demand.
The app's algorithmic recommendations are based on user activity within Sora, IP-derived location data, past post engagement, and ChatGPT conversation history, with an option for users to disable the latter. Parental controls, accessible via ChatGPT, offer features such as overriding infinite scroll limits, disabling algorithmic personalization, and managing direct messaging. The introduction of a social platform necessitates robust user safety protocols, a challenge OpenAI has previously addressed with its ChatGPT platform regarding concerns over access to user likeness and the potential for deceptive content generation. The company acknowledges that non-consensual videos remain a persistent issue within AI-generated media.
