One year after Sora 1 redefined what was possible with moving images, OpenAI has announced the launch of the all-new Sora 2 model, alongside a dedicated social video application, the Sora app. Internally, Sora 1 was viewed as the “GPT-1 moment for video generation,” marking the first time video generation truly felt functional and exhibiting simple behaviors like object permanence. Now, Sora 2 is being hailed as the “most powerful imagination engine ever built,” opening up new possibilities for creativity and connection.
The Sora app, which is currently invite-only, is designed to rival platforms like TikTok, featuring a feed of AI-generated videos. It allows users to create and share strikingly lifelike clips using text prompts. The initial rollout is limited to the iOS App Store in the US and Canada, with an Android version in development.
Sora 2: Features & What’s new?
Sora 2 is OpenAI’s flagship video and audio generation system, representing a significant step-function change in model capability compared to its predecessor. It boasts several key advancements that enhance realism and controllability.
Physics and Motion Accuracy
Sora 2 achieves state-of-the-art results in motion physics, IQ, and body mechanics, marking a giant leap forward in realism. The model is described as being much smarter at physical interactions than any previous video generation system. This improved capability allows Sora 2 to robustly handle complex dynamics and collisions, modeling them in a way that feels extremely natural. Examples of complex movements it can now generate include Olympics gymnastics routines, backflips on a wakeboard, figure skating triple axels, and realistic kick flips.
Integrated Audio Generation
A major new feature in Sora 2 is the simultaneous generation of both video and audio. This is a general-purpose system capable of creating dialogue in various languages spanning multiple speakers, as well as generating sound effects and detailed soundscapes.
Steerability and Narrative Control
The team has also worked extensively to improve the steerability of Sora 2. While prior video generation systems often required a shot-by-shot approach, Sora 2 is significantly better at generating longer narratives that contain multiple shots and tell more coherent stories all in one go.
Furthermore, the model exhibits an incredible dynamic range, avoiding the common issue where prior models collapse into a single aesthetic; Sora 2 can cover anything from high realism to an anime style.
The Cameo Feature: AI Avatars with Explicit Consent
The most unique and celebrated feature of Sora 2 is called Cameo. This feature allows you to step into any generated world or scene and lets friends cast you in theirs. Cameos enable the creation of life-like AI avatars of yourself that can be inserted into any imaginary scenario, such as playing volleyball or wrestling an elephant.
How Cameos Work
The system works by observing a short clip of an individual (or even a pet or object). The model deeply understands the appearance, and can then inject that likeness into any prompt, acting almost like another text token. To set up a Cameo, users must go through a short verification process that includes recording a dynamic audio prompt and passing a “liveness check” by moving their head in specific directions. This validation process is designed to ensure that no one is impersonating you.
Control Over Your Likeness
OpenAI emphasizes the principle of ownership and control over your digital identity. Users have full control of their likeness and must give explicit permission before anyone can generate them. Cameo settings allow you to define who can use your likeness: only you, people you approve, mutuals, or everyone.
If your Cameo is used in a video (even a draft), you are treated as a co-owner of that video, granting you the power to delete it or revoke access at any time. You can also tune Cameo preferences to guide the model on how you wish to be portrayed and mitigate model hallucinations, such as being given a “weird accent or something like that”.
The Sora App: A New Social Experience
The Sora app offers a highly familiar interface, resembling social media platforms with a profile, identity, and the ability to follow others. The core experience is driven by creativity and human connection, despite the content being fully AI-generated.
Encouraging Creation and Connection
OpenAI believes the app’s ease of creation will allow it to lean back into the idea of friend and family connections, which they feel social media has moved away from. The feed will heavily prioritize connected content.
The app features a Remix function that is key to participation in trends and storylines. If you see a video that inspires you, you can click Remix and fire off your own variation instantly. Users can also personalize their viewing experience using a beta feature that allows them to select the type of content they want to see, such as content related to a “relaxing mood” or “Animals”.
The app’s design philosophy seeks to optimize the feed to encourage creativity rather than just passive scrolling. Currently, the app generates videos that are limited to 10 seconds in length.
Safety, Moderation, and Ethical Concerns
The launch of Sora 2 has been met with both excitement and significant ethical concerns, especially regarding the potential for impersonation and framing.
A two-minute celebratory clip met with predominantly negative reactions from netizens, who dismissed it as “unsettling” and “soulless”. The ability to facilitate the AI generation of photorealistic videos raises serious implications for impersonation. Ironically, an OpenAI developer, Gabriel Petersson, demonstrated this capability by generating CCTV footage of CEO Sam Altman “stealing [graphics cards] at Target”.
Critics were quick to point out that this demonstration paints a “dystopian picture” where individuals could easily be framed for crimes they did not commit, especially given past issues with law enforcement using inaccurate AI-powered facial recognition.
Guardrails and Transparency
OpenAI is deploying various measures to ensure responsible use and safety. They have implemented reasoning models to make it extremely difficult to create harmful content on the network. Specifically, it is currently “impossible to generate” X-rated or violent/extreme content, especially concerning the Cameo feature.
Regarding content transparency, all video exported off the app will be visibly watermarked with the Sora animation. The company is also using C2PA standards and internal techniques to trace generations back to Sora if they appear on other networks.
Furthermore, measures are in place to block the depictions of public figures unless they have uploaded a Cameo and provided consent. The company admits that it is starting conservatively with moderation, and users might initially encounter “overblocking”.
How to Get Access
The Sora app launched on iOS in the US and Canada, initially via an invite-based rollout. When users successfully get off the wait list, they receive a push notification and are automatically given four invite codes to share with friends, reinforcing the idea that the app is best experienced in a social way.
Beyond the new social app, the Sora 2 model will also be available on the existing web app, sora.com, which will receive a facelift. OpenAI is also planning to launch an API in the coming weeks, allowing a long tail of use cases where developers can integrate Sora 2 into their own video editors. Creator tools like “Storyboard,” which allows for shot-by-shot control, are also launching soon.
Key Takeaways
- Sora 2 is OpenAI’s next-generation video and audio generation system.
- The Sora app is a new social video platform featuring AI-generated content.
- Cameo is a unique feature allowing users to create AI avatars of themselves.
- OpenAI is implementing safety measures to prevent misuse and ensure ethical content generation.
Join our community by subscribing to our Weekly Newsletter to stay updated on the latest AI updates and technologies, including the tips and how-to guides. (Also, follow us on Instagram @inner_detail for more updates in your feed).
(For more such interesting informational, technology and innovation stuffs, keep reading The Inner Detail).
