The artificial intelligence giant unveiled its new tool on Monday which allows you to create videos from a simple text, an image or another video. But Sora is not yet available in France.
After ChatGPT, it's time for Sora. OpenAI announced Monday, December 9 the arrival of its video generator on screens. This new tool allows you to create a very realistic video, using a description of just a few words given to artificial intelligence. It is available in more than 150 countries, but not in Europe which has significant regulations in this area. Sora, which represents a further step towards the manipulation of reality, is both impressive and risky.
OpenAI, which has already revolutionized the uses of artificial intelligence by launching ChatGPT two years ago, unveiled a preliminary version of Sora last February. At the end of the year, the American AI giant is listing its innovations, such as its new version of Sora, in the form of a mini Advent calendar. “Since unveiling Sora in February, we've built Sora Turbo, a significantly faster version of the model to put in your hands”said Sam Altman, CEO of OpenAI, on Monday.
Sora aims to be a simple tool accessible to the general public. To generate a video, the user only has to give an instruction to the AI by inserting a sentence or an image, what is called a “prompt”. He can then choose the format (for example, 16:9 or square), the quality, the duration (between 5 and 20 seconds) and the number of proposals he wants from AI (up to 4 videos ). In just a few minutes, Sora then offers the user the video(s) it has generated.
There is also the “storyboard” mode, which allows you to create a more complete montage by linking together sequences. To use Sora, you must have a ChatGPT Plus or ChatGPT Pro subscription, at $20 or $200 per month. With ChatGPT Pro, more videos can be generated, in better quality and longer than with ChatGPT Plus.
Sora is available in 160 countries, but it is impossible to connect to it from the European Union, the United Kingdom or even Switzerland. Europe has strict regulations regarding new technologies and the management of personal data. To establish itself there, OpenAI must comply with the GDPR, the General Regulation on the Protection of Personal Data, and the AI Act, the law which aims to regulate the AI sector.
To be authorized, Sora will have to give guarantees on the way in which its users' information is processed. In the meantime, it is possible to circumvent this impossibility by using a VPN, software that allows you to geolocate your connection in another country, such as the United States for example.
Although Sora is not yet available in France, Nicolas Lellouche, journalist at Numerama, was able to test the new tool. “My first impression is very mixed. Out of 15 videos I generated, two are really beautiful and indistinguishable from reality, he reports. The others have very big errors, physical problems, inconsistent details and which allow you to identify in a few seconds that it is false.
“We feel that, even if it is impressive to generate videos in such a short time, we are still far from realism comparable to what we can do with photos today.”
Nicolas Lellouche, journalist at Numeramaat franceinfo
Even if its latest addition does not yet deliver on all its promises, OpenAI has great ambitions with Sora and assures that it will improve over time. “This version is a start, we know it will have errors, it is not perfect”insisted Sam Altman during the presentation of the tool. “For the moment, it's just another tool for generating videos, we are not at all on a tool which will allow people with bad intentions to generate realistic videos of disinformation, retorts Nicolas Lellouche, but it will come. I have no doubt that this type of tool, one day, will be ultra-sophisticated and will be able to deceive people.”
Concerning the fear that Sora allows the production of false information, its designers want to be reassuring. Each Sora video has an electronic signature in its metadata, as well as a small logo embedded at the bottom right of the image to identify that the content comes from an AI, called a “watermark”. “In reality, you just need to recut a video to remove a watermarktempers Nicolas Lellouche. And we know very well that, on the internet, most people will not check the source of a video. So if an image speaks to them and they want to believe it, they will distribute it, before looking in the metadata if it contains an element that says it is false.
More OpenAI a “still took measures”underlines the Numerama journalist. Sora is equipped with a face detector, capable of identifying the presence of a minor, a public figure or a copyrighted element. “For example, if I want to post a photo of myself and ask Sora to turn it into a video, I agree, by checking four boxes, to say that these are indeed images that belong to me. And in the event of a violation copyright, they will know and they can delete my account”explains Nicolas Lellouche.
“We didn’t wait for the arrival of AI to create false images on the Internet, recalls Nicolas Lellouche. What changes with these tools is that it's very simple. On video, it was very complicated, you generally needed very powerful computers to modify a face, etc. There, we are potentially on the verge of being able to do this in 10 seconds with an instruction sentence given to an AI. So where this changes is that we will allow people with bad intentions to do things much more quickly, without technical skills.”
“With each emergence of a new technology, it does good but also harm.”
Nicolas Lellouche, journalist at Numeramaat franceinfo
Although still in its infancy, Sora was a victim of its success from its launch on Monday. “Demand is higher than expected, registrations will be disabled from time to time and generations videos will be slow for a while”wrote Sam Altman on X.
