Samsung AI Deepfakes With Single Photo

First, technology found a way to fake the voices of real people, making them say whatever their creator(s) wanted them to say. Then we saw how Deepfakes could fake a person’s action. After that, we learned that with AI and machine learning, technology can generate realistic-looking people, bring Salvador Dali back to life (granted, this one had an actor involved in the process), and even create non-existence people, complete with body and all and moves too. Now, here’s another proof that technology is getting creepier by the day.

Vice recently reported that researchers at the Samsung AI Center in Moscow have developed a technology that can animate highly realistic portraits with just a few or, in some instances, a single image. The results, if you are going to be picky, are not mind-blowing given the amount of artifacts the results bear. They are nevertheless convincing enough and suitably creepy. If down with learning the details, you can read the paper, Few-Shot Adversarial Learning of Realistic Neural Talking Head Models, published on the preprint server arXiv.

Obviously, if one or few images could be so realistic, you can only imagine realism if they are more resources (known as training examples) available. But what really makes this technology so impressive is, it does not simply cut and paste a face, and learning from a catalog of expressions like Deepfakes and generative adversarial networks does. Instead, Few-Shot Adversarial Learning uses facial features common across humans to manipulate the new face. In other words, it learns the appearance of the person and using what it knows about human expressions and blends the two together.

NOW READ  Yomiplanet x FoloToy Sentry Frank W. Murphy AI 12-inch Doll: World’s First AI Doll

What’s even more mind-blowing is, it is able to generate the talking head in a new view angle even if the angle is not present in the training examples. While at this point, the results have visible stitching and artifacts, it is a matter of time before this technology is perfected and when that happens, we will be truly wondering what is real and what’s not every.single.time. If you ask me, this development is both to be celebrated and cheered and also to be wary about at the same time. Here, have a look at the video of this new development in action.

Images: YouTube (Egor Zakharov).

Source: Vice.