Digital Economy

OpenAi launches the Sora app and invades the social field of TikTok and challenges YouTube

The Sora 2 has already been launched, with the iOS app available by invitation only in the US and Canada.

by Luca Tremolada

1 October 2025

3' min read

Translated by AI

Versione italiana

3' min read

Translated by AI

Versione italiana

OpenAI presented Sora 2, an artificial intelligence model that generates videos from textual instructions, and a dedicated app for iOS. This is not just a technical update on the first Sora. It is a change of strategy. Sam Altman's company is entering directly into the social video terrain, the same terrain where TikTok, YouTube and Instagram operate. The difference is that content is not uploaded by users but generated within the app.

How it works.

The model promises more realism. Fewer physical errors, more consistent movements and synchronised audio. OpenAI speaks of a wider range of styles, from photorealistic to animation. Each video is branded with watermarks and hidden metadata. There is also a 'cameo' system: users can decide who can use their face or voice in videos and revoke consent at any time.

Tecnicaemente Sora 2 functions as a large multimodal transformation model: it takes a text or visual prompt and translates it into a sequence of frames with coherent audio. It does not build the videos frame by frame in a static manner, but generates them as if it were simulating a continuous scene with actors, objects and movements. This is the difference from Sora 1, which often struggled to maintain temporal coherence: a glass might change shape between frames, a character suddenly disappear, movements follow unphysical logic.

With Sora 2, OpenAI aimed at better modelling of the laws of the world. The system learns from large amounts of data not only to imitate the visual appearance but also the dynamics of events. For example, if a dog runs towards the water, the model tries to represent the impact with the surface and the resulting waves in a credible way. Another new feature is the management of audio: no more artificial soundtrack, but lip-synchronisation and background noise linked to the scene.

The real difference, however, lies in the app. The first Sora was a tool for those who wanted to experiment with generative AI, without a social context. Sora 2 becomes an ecosystem where videos are created, shared and edited directly within a feed. OpenAI does not just want to provide the technology, but to create the platform where technology becomes a cultural product.

What are cameos? And what is consensual deepfake?

To avoid privacy violations and deepfakes OpenAi invented cameos. Basically yu register yourself (face and voice) inside the app, giving explicit consent. From then on, the system can use your 'realistic avatar' in the generated videos: you can appear in scenes you have never shot, speak in your own synthesised voice, or interact in a fantasy context (like dancing on Mars or acting in a horror trailer). If someone else wants to use your cameo, they must have your permission: the app stipulates that videos are "co-owned" between the person who created the prompt and the person who lent their face/voice. You can revoke permission: if you no longer want to, you can block the use of your cameo, and videos that include you are removed or blocked. It's basically a consensual deepfake system: instead of stealing someone's face, each user decides whether they want to be 'included' in the generation. The difference from Snapchat's filters or TikTok's duos is that here it's not just about funny overlays, but full generation of content in which your digital alter ego can play entire roles.

TikTok and YouTube?

The competition is direct. TikTok and YouTube Shorts have billions of users and recommendation algorithms that keep people hooked. Meta and Google are experimenting with similar generative models, but integrating them into existing platforms. OpenAI starts from scratch. It has no previous content base. It has to convince users and creators to build an entirely new ecosystem.

The risk is obvious. If the quality of the videos does not cross the threshold of credibility, Sora will remain a technological curiosity. If, on the other hand, it manages to guarantee speed, realism and control, it can become a social laboratory alternative to traditional feeds. Altman himself recognises this. Social media have positive and negative effects. This is why OpenAI tries to draw barriers from the outset: content filters, usage limits, mandatory watermarks.

The game is open. On one side are giants with ready-made audiences and infrastructure. On the other a model that brings in the idea of 'native AI-generated videos'. For the online content economy, it means testing whether there is a market where videos are no longer documents of the real thing but synthetic products, shared and remixed as if they were digital Lego bricks.

Luca TremoladaGiornalista
- @lucatremolada
- LinkedIn
Luogo: Milano via Monte Rosa 91
Lingue parlate: Inglese, Francese
Argomenti: Tecnologia, scienza, finanza, startup, dati
Premi: Premio Gabriele Lanfredini sull’informazione; Premio giornalistico State Street, categoria "Innovation"; DStars 2019, categoria journalism
Scheda autore
Trust project