Artificial intelligence creates realistic video from a single photo

Samsung engineers have improved the artificial intelligence that creates the so-called “artificial intelligence”. deepfake and have taken it to a whole new level. Just one photo is enough for algorithms to create realistic video based on it. The authors demonstrated their technology using the example of a portrait of the Mona Lisa. Effect? The Mona Lisa moves, speaks, blinks its eyes, and it all looks disturbingly good.

Deepfake is a technique for creating or modifying materialoin video using an algorithmoin artificial intelligence. The name comes from a combination of wordsoin deep learning – one of the machine learning methods based on artificial neural networks – and fake, which in English means counterfeit, false, subrobka. Technology makes it possible to generate materialoin video, whichore show something that didn’t actually happen.

Artificial intelligence’s capabilities in generating deepfake are growing and are also of growing concern. The performance of the engineeroat Samsung show that a single photo downloaded from at least a social network is enough to create fake video footage. It’s not hard to imagine what such a tool can be used for in the hands of a crook.

Recent improvements to the algorithmoin artificial intelligence in generating deepfake content brings this technology to a whole new level and is the work of an engineeroIn from Samsung’s research lab in Moscow. The new algorithms can not only operate from a smaller initial database of imageow (it only takes one), but roalso create videos in a very krotkim time.

„Our technology will have applications in video conferencing or video games, but also in the effects industryoin special” – wrote tworcy in an article published in the pre-print archiveoin The technology will also have enormous potential for abuse.

Samsung’s algorithms were trained on a public database of 7,000 celebrity photosow. The system works based on a number of features thatothat can be manipulated – characteristic facial features or facial expressions. The system mimics these features and gives them realistic-looking movements. With training, it is able to match facial expressions to the appearance of the character in the photo, resulting in realistic video footage.

Artificial intelligence has generated fake videos of the Mona Lisa, Albert Einstein, Salvador Dali, Fyodor Dostoevsky or Marilyn Monroe. The effect can be seen below.

Admittedly, videos generated from a single photo are not as good in quality as, say, those created using a dozen or more photographs. But they may be enough to fool someone. On top of that, the technology is developing at a rapid pace, also it is to be expected that wkrotce quality will improve significantly.