Brief Summary
The video introduces Sora 2 and the Sora app, highlighting advancements in video and audio generation, physics, and realism. A key feature is "Cameo," allowing users to insert themselves or others into generated scenes. The Sora app aims to provide a new, creative social media experience centered around AI-generated content, emphasizing user control, safety, and connection with friends.
- Sora 2 introduces audio generation and improved physical interactions.
- The Cameo feature allows users to insert themselves or others into AI-generated videos.
- The Sora app is launching on iOS in the US and Canada with an invite-based system.
Introduction to Sora 2 and the Sora App
OpenAI announces Sora 2 and the Sora app, marking a significant advancement in AI-driven video generation. Sora 2 is described as a powerful "imagination engine" with new features, including sound generation. The improvements in motion, physics, image quality (IQ), and body mechanics contribute to enhanced realism. The introduction of the Cameo feature allows users to insert themselves into any generated scene, offering new creative possibilities.
Key Features of Sora 2
Bill, the head of Sora, explains that Sora 2 is a flagship video and audio generation system. It excels in physical interactions, handling complex dynamics like gymnastics routines and wakeboarding backflips with greater robustness and naturalness. Sora 2 also offers improved steerability, enabling the creation of longer, more coherent narratives within a single generation. A major addition is audio generation, allowing the model to simultaneously create video and audio, including dialogue in multiple languages, sound effects, and soundscapes.
The Cameo Feature Explained
The Cameo feature is unique to Sora 2, enabling users to insert themselves or others into Sora-generated environments by observing a short video clip. This capability stems from the model's world simulation, allowing it to understand and integrate any observed clip—be it a human, pet, or object—into a prompt as a text token. This feature is seen as a new form of communication, evolving from text messages and emojis to a video-based medium.
Introducing the Sora App
Rohan, who leads the Sora product team, introduces the Sora app, designed to capture the capabilities of Sora 2. The app features a familiar social media interface with profiles and the ability to follow others, but all content is AI-generated by humans. The app is currently pre-revenue. The feed showcases the Sora team's creations, including memes and videos demonstrating the model's capabilities, such as dynamic range and stylistic versatility.
Demonstration of the Sora App Features
Rohan demonstrates the Cameo feature within the app, showing examples of himself and Sam in the same scene, highlighting realistic details like shot changes, natural gestures, facial expressions, and accurate lip-sync. He also showcases the model's ability to render pets in different styles, such as anime. The app includes a simple composer where users can describe their ideas and generate videos, incorporating cameos of themselves or approved friends.
Cameo Setup, Permissions, and Safety
Rohan explains the Cameo setup process, which involves recording a dynamic audio prompt and undergoing a liveness check to prevent impersonation. Users have full control over who can use their cameo, with options to allow only themselves, approved individuals, mutuals, or everyone. Users can also guide the model on how they want to be portrayed through Cameo preferences. Ownership of content created with a user's cameo, with permission, belongs to the user, allowing them to delete it.
Remix Feature and Content Examples
The remix feature allows users to create their own variations of existing content, participating in trends and storylines. Examples include remixes of a Sora-themed perfume ad, turning it into ads for toothpaste and other products. The feed also showcases the model's physics capabilities, such as a coworker performing a kickflip, and humorous content featuring cameos with gold chains and other fun elements.
Philosophy Behind the Sora App
Thomas, who leads Sora engineering, shares initial skepticism about an AI-generated feed but notes that the Cameo feature creates a new way of connecting with friends. The app prioritizes connected content and offers features to control the feed, such as selecting the type of content to be displayed based on the user's mood. The goal is to inspire creativity and encourage users to participate actively rather than passively scrolling.
Safety, Moderation, and Future Services
Rohan discusses safety and moderation measures, including separate policies for users under 13, clear labeling of AI-generated content, and provenance techniques like watermarks and internal tracing. Reasoning models are used to prevent the creation of harmful content, especially within the Cameo feature. The team is starting with conservative moderation and is working to balance user freedom with safety. Sora 1 on sora.com will also receive the new model and features like storyboard, and an API will be launched for integration into other video editors.
Launch Details and Closing Remarks
The Sora iOS app will be available for download in the App Store later in the day, initially in the US and Canada, with an invite-based rollout. Users will receive four invite codes to share with friends. The Sora research program aims to build AI systems that deeply understand the physical world, and the app is intended to bring joy and creativity to users. The team expresses excitement to see what users will create on the Sora app.
