The artificial intelligence of Sorathe OpenAI model capable of generating realistic high-resolution videos or extending existing videos starting from simple text commands, is officially available for those who have a subscription to ChatGPT Plus And Proeven if not in EU countries and therefore not in Italy. The announcement came yesterday, Monday 9 December 2024. The release of Sora came after years of research and development, culminating in the release of the version “Sora Turbo”which offers significantly improved performance compared to previous prototypes. With this release, users can generate videos up to 1080pOf maximum duration 20 secondschoosing between widescreen, vertical or square formats. Despite having great potential, it must be said that Sora still presents some technical and, to be honest, also ethical limitations, given its (potentially devastating) impact on the film and art industry and the implications in terms of privacy and security that it affects all of us.
What Sora can do: functions and features
Sora is capable of generate realistic videos starting from text. Users can combine existing assets such as images and clips, or generate completely new content, thanks to advanced tools such as storyboards. This allows precise control over each frame, making it easier toanimation of static images (such as photos and illustrations), theextension of pre-existing videos (which may be missing some clips) and the creation of videos through the remixes of various contents.
There “Turbo” version Furthermore, Sora is able to make the entire content generation process faster and more efficient, reducing waiting times and improving interaction with the interface. Regarding Sora Turbo, in fact, OpenAI declared that it is «significantly faster than the model previewed in February».
It must be said, however, that Sora’s algorithm is not perfect: the physics generated in the videos can sometimes appear unrealistic or, in any case, unnatural. OpenAI said it will continue working to refine the model and make it accessible to a broader audience. In any case, if you want to get an idea of what Sora is capable of, take a look at the following video.
What are Sora’s subscription plans
At the time of writing this article, Sora is included in the plans ChatGPT Plus subscription And Pro. These are the characteristics of both plans.
- ChatGPT Plus: starting from $20/monththis plan allows you to generate 50 videos per month of a maximum duration of 20 seconds at 720p resolution.
- ChatGPT Pro: starting from $229/monththis plan offers the ability to generate unlimited videos up to 1080p and also allows you to download the content produced without a watermark.
OpenAI has made it known, however, that it is also working on development of a free version of Sorawhich will see the light in the future.
Because Sora is not available in Italy
There Sora’s availability is closely linked to data regulation and compliance with local laws. Currently, the service is accessible in all countries where ChatGPT is available, except for the United Kingdom, Switzerland and the European Union. In this regard, the company that develops Sora has specified:
Europe is a key market for OpenAI as it continues to demonstrate its commitment to working alongside EU and UK governments and regulators to make its products available in these regions.
Although not available in Italy, it would seem that Sora is still available in the Vatican City (which for the record is not part of the EU), as documented in the following screenshot.
Privacy and security: what are the risks of Sora?
Since there is an “elephant in the room” – i.e. the conversation about possible privacy and security risks of Sora – let’s conclude the in-depth analysis by talking about these. Let’s start by saying that OpenAI has tested the service in recent months with the help of external experts. This led to marking each generated video with a visible watermark and C2PA certified metadatathanks to which it is possible to trace the artificial origin of the contents shown in the videos. They are also present security filters to mitigate the deepfake phenomenon e image classifiers work to review the frames of each generated content ensuring that it adheres to OpenAI’s usage policies before it is displayed to the user.
Regarding the issue of deepfakes, OpenAI also stated:
The “likeness” feature is currently only available as a pilot feature for a small group of early testers. To address concerns about likeness misappropriation and deepfakes, we have set particularly strict moderation standards for uploads featuring people and continue to block content with nudity. Users who have access to this feature will also see reminders in the product about our policies, including what is allowed and what is not. We will actively monitor patterns of misuse, and when we find them, we will remove the content, take appropriate action, and use these early learnings to iterate our security approach.
Whether these measures will be enough to contain the privacy and security problems inherent in AI models for generating videos, such as Sora, will be told by the future.
