Games

Copyright, VIPs and privacy: our test of the Sora 2 app

-OpenAi's image generator is not yet available in Italy but is already causing much discussion

by Alessandro Longo

5' min read

Translated by AI
Versione italiana

5' min read

Translated by AI
Versione italiana

Entri e l’effetto è straniante. L’app ti avvisa subito, però. E ci sembra di assumere i panni di Neo in Matrix: “le facce sembrano richiamare persone vere ma nulla è reale”, dice l’avviso. Già perché l’app Sora di OpenAI, che abbiamo provato anche se ufficialmente è disponibile solo negli Usa e in Canada, è proprio questo: il primo caso di piattaforma che rende l’irreale virale, popolare. Come TikTok ma tutta fatta con video (di dieci secondi) creati con l’intelligenza artificiale. Tra i più popolari c’è un video dove una scimmia impazzita cerca di entrare in un aereo. Sullo sfondo, un tornado, che poi la risucchia. La visuale è in soggettiva di una persona seduta nell’aereo e che guarda la scimmia. Inquietante: prima dell’AI generativa sarebbe stato necessario essere esperti di computer graphics per poter fare un video del genere. Adesso basta digitare qualche parola nel box dell’app e aspettare circa un minuto.

Sora 2, la nostra prova/1

It must be because of this surprising novelty that Sora surpassed one million downloads within seven days of its launch. It was faster than Chatgpt.

Loading...

To try it out, you have to have an invitation code offered (for free) by another user. You can use it via app (iPhone only for now) or browser. And be in the US, Canada or pretend to be there: we used it via Opera's browser, which has a free vpn in it.

When you log in, 'For you' videos (like TikTok) appear, but you can also see the most popular or the latest ones and those of people you follow. A special feature is the cameo function, which allows you to upload a short clip of yourself to be automatically included in the generated clips.

It is not necessary to have a cameo to generate a video, however. In the Describe your video box we can freely describe our video and possibly even match it with an image that the AI can use. Sora adds the audio, even in Italian (if we ask to use this language).

Sora 2, la nostra prova/2

It does this even if we don't ask for it. For example, we asked it to generate an image with two pigeons and a chick, without adding any images, and the app did it: a very realistic video, with the audio of a person commenting on it in an excited tone ("look, the chick is less than the size of a hand!"). At its base is OpenAI's new AI video generator Sora 2. We can also download the videos, which can then end up on other platforms and social networks.

We tried uploading photos of Giorgia Meloni in Gaza or asking to generate a video with her appearance and the app blocked us for violating the terms of use. It happens a lot: now on Sora there is a black and white video with Martin Luther King dreaming of a future in which the app will not block users' videos so often.

OpenAI is trying to curb a phenomenon that is still partly out of control: the improper use of the faces of celebrities or deceased persons in fake or disparaging contexts. Videos depicting, for instance, Stephen Hawking in surreal or offensive situations. OpenAI has reacted by introducing stricter limits, such as blocking copyrighted characters at the request of rights holders. Let us remember that the possibility of exporting videos from Sora means that the phenomenon can overflow into any social network, which is already very rich in AI-generated content. The success of Sora could therefore make deepfakes an unstoppable phenomenon on social networks.

The biggest problems arose with cameos, which are 'reusable characters' modelled on users based on the audios and videos they upload. Users must consent to the creation of their own deepfake and can then grant access to their digital image on four levels: only you, people approved by you, friends or everyone. That is: our image could be used for any purpose. A few hours ago, however, in a post on X, Sora boss Bill Peebles announced the possibility of blocking our Cameo from the settings.

Users can set more precise limits on what their Cameo can do and say using text prompts, such as 'Don't put me in videos that contain political comments' or 'Don't make me say this word'. One can also ensure that the Cameo appears with specific details, such as wearing an identifying garment.

Il contesto normativo italiano rende il caso Sora particolarmente interessante. L’Italia è tra i primi Paesi europei ad aver approvato una legge specifica contro i deepfake, entrata in vigore il 10 ottobre scorso, che punisce con la reclusione da uno a cinque anni chi diffonde, senza consenso, immagini, audio o video generati o manipolati con intelligenza artificiale in modo tale da indurre in errore sulla loro autenticità e arrecare un danno ingiusto. Si tratta di una norma costruita per colmare un vuoto giuridico: fino a poco tempo fa, i casi di falsificazione digitale venivano inquadrati come diffamazione, trattamento illecito di dati personali o violazione del diritto all’immagine. La nuova disciplina, invece, affronta direttamente il problema del realismo crescente delle tecnologie generative e della loro capacità di ingannare l’occhio umano. Sul versante del copyright, poi, la situazione è altrettanto complessa. In Italia, il diritto d’autore tutela non solo le opere artistiche ma anche l’

Sora 2, la nostra prova/3

Italian and European legislation, then, obliges the use of AI in videos, images, text or other generated content to be made transparent.

Sora's videos in fact have digital watermarks in the videos and automatic prompt control systems to prevent the creation of banned content. However, users have already learnt how to remove watermarks and there are always new ways to circumvent bans on deepfakes and copyright violations. Among the most popular videos is one of a woman crying with a pig on a plane, similar in appearance to the girl who wanted to breastfeed a cat on a flight and about which there are many memes online. Then there are the Star Wars-themed videos.

We tried to upload a caricature of Giorgia Meloni (which would be copyrighted by the cartoonist and the publisher) and Sora took it. He put it in a video that also depicted a woman, similar to Giorgia Meloni (thus inspired by the caricature), against the backdrop of Gaza City in ruins. And he made her say a few words about the disaster.

For now, despite the immediate success, there are not many users and videos on Sora. Who knows what will happen, with deepfakes of all kinds, when the global phenomenon breaks out.

Copyright reserved ©

Brand connect

Loading...

Newsletter

Notizie e approfondimenti sugli avvenimenti politici, economici e finanziari.

Iscriviti