Deepfake videos have been making their rounds on the internet, and safe to say that they are deeply disturbing at how realistic some of them are. However, it seems that Samsung has taken the technique involved and making it even better where they have developed an algorithm that can actually create an entire animated video of you using just a single photograph.
Dubbed “Few-Shot Adversarial Learning of Realistic Neural Talking Head Models”, the paper was published by researchers to arXiv.org where if you’re interested in learning more, you can take a look at it. The paper shares the researchers’ findings, where they were actually able to take a single image, like the painting of the Mona Lisa, and animate it where it looks like the painting is talking.
To achieve this, the AI was taught how to identify what the researchers call “landmark” features of the face, such as the eyes, mouth, length, and shape of the nose bridge. According to the researchers, they believe that this tech could be used for gaming, special effects, or avatars in video conferences.
While we do acknowledge those uses, at the same time, we imagine that it could create a lot of misinformation and confusion when used for more malicious purposes, such as making a video of someone saying something that they did not. You can check out the video above to see the very realistic application of this algorithm.