Samsung is developing an algorithm that only needs one image to make a fake video

Samsung is developing an algorithm that can make fake talking-head videos with just ONE photo

  • A new algorithm can create fictitious videos with just a single image
  • The technology uses & # 39; landmark & ​​# 39; functions to bring images to life
  • In a video, the algorithm animates Salvador Dali, the Mona Lisa and more
  • Some have made rapid advances in technology

As if the world of fictitious photos & video was not scary enough, researchers from Samsung & # 39; s AI center in Moscow demonstrated an algorithm that can make video & # 39; s with just one image.

In a video demonstration and a paper published in the pre-print journal ArXiv, the researchers show the possibilities of what is described as & # 39; one-shot & # 39; and & # 39; few-shot & # 39; machine learning.

The results of their system bring popular faces to life such as those of surrealist painter Salvador Dali and actress Marilyn Monroe with a single photo.

Scroll down for video

The latest algorithm was trained using the publicly available VoxCeleb database that contains more than 7,000 images of celebrities from YouTube videos & # 39; s. With the help of that dataset, the AI ​​was able to determine what researchers & # 39; landmark & ​​# 39; characteristics

The latest algorithm was trained using the publicly available VoxCeleb database that contains more than 7,000 images of celebrities from YouTube videos & # 39; s. With the help of that dataset, the AI ​​was able to determine what researchers & # 39; landmark & ​​# 39; characteristics

The latest algorithm was trained using the publicly available VoxCeleb database that contains more than 7,000 images of celebrities from YouTube videos & # 39; s. With the help of that dataset, the AI ​​was able to determine what researchers & # 39; landmark & ​​# 39; characteristics

The more images are entered into the program, the more realistic the final video becomes.

Although a single image translated into a moving face may be noticeably altered, an example of 32 images produces a moving image with near-to-life accuracy.

The technology of researchers seems to use algorithms that are used to produce deepflies that far exceed existing methods, not only in terms of realism, but also in the amount of work required to create photorealistic forgeries.

Current AI systems are usually trained in an extensive catalog, which means that the algorithm must scan large sets of data about the body or face of a subject.

This method limits the application of the system to people who have sufficient photos and videos available for use.

In the case of the new software researchers, however, those data sets are more of a luxury than a necessity.

The latest algorithm was trained using the publicly available VoxCeleb database that contains more than 7,000 images of celebrities from YouTube videos & # 39; s.

With the help of that dataset, the AI ​​was able to determine what researchers call & # 39; landmark & ​​# 39; characteristics – universally identifiable properties – between noise, eyes and more of subjects.

Because the algorithm recognizes common features as opposed to specific features of a subject, it can quickly extrapolate images with little input and, in some cases, bring still lifes to life of people like Fyodor Dostoevsky, who died in 1881 and probably never was captured on video.

This method also means that the technology applies to non-celebrities.

Using just a few selfies from a topic, a demonstration shows how the algorithm can create a video facsimile to simulate how they can move their heads and speak.

The AI ​​can currently only create & # 39; talking head & # 39; style video & # 39; s subjects from the shoulders, but if the algorithm is similar to similar freezing systems, that can change quickly

The AI ​​can currently only create & # 39; talking head & # 39; style video & # 39; s subjects from the shoulders, but if the algorithm is similar to similar freezing systems, that can change quickly

The AI ​​can currently only create & # 39; talking head & # 39; style video & # 39; s subjects from the shoulders, but if the algorithm is similar to similar freezing systems, that can change quickly

In a video demonstration and a paper published in the pre-print journal ArXiv, the researchers show the possibilities of what is described as & # 39; one-shot & # 39; and & # 39; few-shot & # 39; machine learning

In a video demonstration and a paper published in the pre-print journal ArXiv, the researchers show the possibilities of what is described as & # 39; one-shot & # 39; and & # 39; few-shot & # 39; machine learning

In a video demonstration and a paper published in the pre-print journal ArXiv, the researchers show the possibilities of what is described as & # 39; one-shot & # 39; and & # 39; few-shot & # 39; machine learning

The AI ​​can currently only create & # 39; talking head & # 39; style video & # 39; s subjects from the shoulders, but if the algorithm is similar to similar freezing systems, that can change quickly.

Another recent and surprising technology recently demonstrated by former Google and Microsoft employees shows how an algorithm can convincingly replicate the voice of popular podcaster, Joe Rogan.

Deepfake technology skeptics say these two technologies together enable bad actors to make fake videos in a practical way that can spread misinformation about politicians or steal the identity of other victims.

Deep-eating technology has long been used to place celebrities' faces on the bodies of porn actors, calling some people into legislation that would prohibit its use.

WHY IS FALSE CELEBRITY PORN MADE BY AN AI SO CONCERNED?

In December, it was discovered that Reddit users created fake pornography with celebrity faces stuck to the bodies of adult film actresses.

The troubling videos created by Reddit user deepfakes look strikingly real as a result of an advanced machine-learning algorithm that uses photos to create human masks that are then overlapped on top of film recordings for adults.

Now AI-assisted porn is spreading everywhere in Reddit, thanks to an easy-to-use app that can be downloaded directly to your computer, according to on motherboard.

The video, made by Reddit user deepfakes, shows a woman taking over the rough resemblance of Gadot, with the actor's face over the head of another person. A clip of the video is displayed

To create the similarity of Gal Gadot, for example, the algorithm was trained in real porn videos & images of actors, allowing it to make an approximation of the actor's face that can be applied to the moving figure in the video.

Since this is all freely available information, it could be done without the permission of that person.

And as Motherboard points out, people nowadays constantly upload photos of themselves to various social media platforms, meaning that someone could use such a technique to harass someone they know.

. [TagsToTranslate] Dailymail