Deepfakes Generated In Real-Time By System With New Algorithm

In 2017, a technique to alter a person’s appearance on images or videos with the face of somebody else using artificial intelligence has been born with the title “deepfake”. Recently, a newly-developed system named FSGAN has the ability to carry out this technique in real-time. In other words, the algorithm can change people’s faces without all the comprehensive instruction that a normal AI would need to take in about a particular face before performing any changes.

As stated by Motherboard, this ground-breaking improvement means that the amount of deepfake contents will grow at a greater speed than we have ever seen before, as now such manipulated and deceitful clips can easily be produced with so little technical knowledge.

FSGAN was built by a team of researchers from Bar-Ilan University in Israel. This system functions by analyzing the facial movements of a certain someone in a specific video. Afterward, those expressions will be matched to a different person’s face for superimposition. After conducting research, ArXiv concluded that any deepfake creator would only need a picture of their target in order to create whatever fraudulent content they wish to.

In the example video below, it is not hard to see the imperfections in the results. The background was inconsistent at times, and some pairs of faces swapped – such as Regina King with the face of Alfonso Cuaron at 1:13 – is clearly for the purpose of entertainment. Nevertheless, this video has shown the possibility of this deepfake technique being used to create credible contents to fool the minds of the audience.

According to Motherboard, the developers of this unprecedented technology stated that their works would be published and accessible to the whole world. In their opinion, other similar systems would still come to exist even if they kept their findings in secret.

That reasoning is actually common among deepfake generators. They claim to share their technology so that someone, but clearly not them, has the opportunity to look into it and come up with some effective countermeasures to said technology. In reality, it would be much simpler if, for example, such inviting, eye-catching deepfakes were not created at all.