Open your phone, record a few seconds of your face and voice and, instantly, see yourself in a scene that didn't exist a while ago: that's the new promise of OpenAI with Gravel 2. The application makes you the protagonist of a clip that keeps movements, rebounds and continuity as if it had been filmed with a camera.
The idea isn't particularly cryptic: you enter your identity once, and the app inserts you into AI-generated scenarios. What's striking is the sense of creative control and the system's ability to chaining shots without the “world” falling apart along the way. Hence, we speak of “OpenAI's TikTok".
What is Sora 2 and why is it called OpenAI's TikTok?
Sora 2 is the next generation of OpenAI's video model, launched alongside a social app called Sora. The formula combines a clip generator with a TikTok-style feed, remixes, and options to use your own image as a cameo.
The social approach doesn't stop at scrolling. OpenAI offers an experience designed to create with other people: you can remix other people's videos, respond with variations and maintain creative threads around the same scene or idea.
The app is available in a limited way: only iOS, by invitation and in the United States and Canada For now, the company is talking about a gradual expansion and web access for those who receive an invitation, but with no fixed dates for other countries.
In parallel, the Sora 2 model also positions itself against alternatives such as Runway (Gen-4) or Google Veo: Its unique advantage is that it combines technical power and social distribution in a single product.
Model Updates: Physics, Continuity, and Integrated Audio
OpenAI equates Sora 2's jump to what GPT-3.5 was once for the text. The big difference is in the physical fidelity: if a ball doesn't go in, it bounces off the board instead of "teleporting," a typical flaw in previous generations.
The model maintains the “state of the world” between shots, which is key to telling short stories in multiple planes. This continuity allows transitions and scene changes without losing characters or props along the way.
In addition to video, Sora 2 generates synchronized audio: voices, effects, and soundscapes that match what's happening on screen. The image-sound mix is born from the prompt itself, simplifying the workflow for creators.
The resulting clips can span realistic, cinematic, or action styles. anime, and for now they move in short durations, of about 5 to 10 seconds, enough for viral pieces and micro-narratives.
OpenAI admits that there are still errors and that consistency is not perfect, but maintains that the progress compared to Sora 1 is notable, especially in stiffness, buoyancy and contact between objects and people.
The Sora app: cameos, remixes, and a language-driven feed
The most distinctive feature is the cameos: After a brief facial and voice verification, you can appear in any generated scene. It's opt-in; you decide if your image can be used and who can use it.
The feed incorporates a “new class” of recommender that accepts instructions in natural languageBy default, it prioritizes people you interact with and shows you content that the system believes is more likely to inspire you to create rather than just watch.
The app makes remixing easy: you take another user's clip, add text changes, and create your own version. This mechanism aims to boost the co-creation and chain variations on the same prompt or scenario.
To strengthen authorship and traceability, OpenAI includes metadata and watermarks indicating that the material was generated with AI, and limits capture with measures such as blocking screen recording.
In the near future, the company is talking about opening the access via API, keep Sora 1 Turbo and offer more quality control in a Pro variant tied to ChatGPT Pro subscriptions.
Well-being, safety and boundaries: policies and moderation
OpenAI claims it does not optimize for continuous usage time: the design is geared toward creation versus consumptionThe app periodically polls you about your well-being and offers to adjust your feed with clear options.
For minors, default limits apply to the number of visible generations, stricter permissions on cameos and parental controls from ChatGPT to adjust personalization, timing, and direct messages.
The company is strengthening human moderation to respond quickly to cases of harassment or misuse. It also indicates that public figures without consent (cameo uploaded by the person) nor adult or extreme content.
- Mandatory verification of face and voice to activate cameos.
- Full user control over who can use your cameo, with the option to revoke access.
- Signs of origin: metadata and watermarks in the clips.
- Measures against abuse: human moderation and limits for teenagers.
In terms of availability and cost, Sora 2 starts off free with limits tied to computing power. ChatGPT Pro subscribers have access to Sora 2 Pro with higher quality, and a wider rollout is expected as the infrastructure grows.
For now, Sora 2 aims to turn short videos into a shared gaming space, where anyone can star in scenes in seconds and share them on a controlled social environmentIf it succeeds, it could mark a period of more participatory creativity, with pieces that blend physical realism, narrative continuity, and sound generated without leaving a mobile phone.
