Pornographic content accounts for 96% of all deepfake videos circulating on the Internet. This is one of the discoveries made by the study The State of Deepfakes, developed by security firm Deep Trace, which specializes in tracking materials of the type created with artificial intelligence to protect the victims’ reputation.
READ: Fake porn videos use celebrities and generate controversy; Look
According to the survey published last week, there are currently 14,678 videos that falsify the characters’ identities in some way, a number that almost doubled in less than a year. The technique has already been used for videos in which former President of the United States, Barack Obama, appears cursing the current Donald Trump, in addition to pornographic montages with actresses Gal Gadot and Emma Watson, for example. Next, see these and other curious facts revealed by the research.
Actress Gal Gadot was the victim of deepfake videos – Photo: Pond5
Want to buy cell phones, TV and other discounted products? Discover Compare TechTudo
1. 96% are pornographic
The study points out that videos with deepfake have been bringing more and more varied themes, but there is still a strong dominance of pornography. Such content accumulates more than 134 million views on adult websites – of the ten largest porn portals, eight feature content with deepfakes. Most of the material, however, is hosted on nine sites dedicated exclusively to deepfakes.
2. Types of deepfake: faceswap, deepnude and lipsync
The report identifies three main types of deepfake in the videos that circulate on the web. The most common and prevalent in pornographic videos is faceswap, which replaces someone’s face with another face, usually that of a famous person. Lipsync, on the other hand, is more present in satire videos and does not involve changing the face, but interferes with the way the subject’s mouth moves to make it look like it says something different than in the original video.
There is also deepnude, a modification that makes a person’s clothes disappear and creates a fake video with them completely naked. The difference with faceswap is that, instead of transplanting the face to someone else, the victim’s own body is changed by the software.
Deepfake with lipsync makes the victim say anything on video – Photo: Disclosure / Deep Trace
3. Women are the main target
Another fact pointed out by the study is the gender cut of deepfakes: the victims of pornographic videos are always women, whose faces are usually placed on bodies of actresses to look like they are acting in a porn movie. On the other hand, videos with deepfake that are published on YouTube, most of the time (61%), target men. In the general scenario, however, women are the most recurrent target.
Women are victims of all pornographic deepfakes, which account for 96% of the total – Photo: Disclosure / Deep Trace
4. United States and South Korea lead
Of all the people who appear in the videos with deepfake, the majority are actresses from the United States or K-Pop stars, a very popular rhythm in South Korea. North Americans account for 73% of the content with deepfake without porn and 41% of pornographic videos – in adult material, another 25% stay with South Koreans. Brazilian personalities do not appear in the survey.
5. Actress with 257 videos is the biggest victim
Entertainment stars are victims in 99% of porn videos and 81% of other types of deepfake distributed online. The appearance champion is a British actress who has accumulated 257 different videos and 5.8 million views. Second place is a South Korean music artist whose face appears in 127 adult videos. The Deep Trace study does not reveal the names of the victims. In the case of the Israeli actress who appears in eighth place, it is easy to conclude that the victim is Gal Gadot, known for playing Wonder Woman and who has become one of the most famous cases of deepfake.
Deepfake victims are almost always actresses and female artists – Photo: Disclosure / Deep Trace
6. GAN technique is the most popular
There are several techniques for producing a video using deepfake. According to the survey, the most popular is GAN, which stands for Generative Adversarial Network, famous for the variety of possible applications and realistic results. One of the clues to the high demand for the method is the growing number of scientific articles published on the topic: since its invention in 2014, works that mention the technique have increased in number exponentially and already registered 1207 publications in 2019 alone.
7. Deepfake appears in satires, scams and political attacks
Outside the pornographic circle, deepfake videos grow especially in three areas. One is satire content, commonly involving celebrities, media professionals and politicians. There are also cases of scams with deep learning software: in September, a German company suffered a millionaire loss after a criminal used an artificial intelligence system to imitate the CEO’s voice and order transfers.
The study does not yet record cases of deepfake in elections, but the technology is already being used as an excuse to attack rivals in the political world. One of the most prominent cases happened in Malaysia: in June, an erotic video of a government member leaked on the web and caused a crisis, but the subject claimed to have been the target of deepfake fraud. According to the report, however, international experts did not find any signs of forgery in the material.
Is Doria’s intimate video deepfake or real? Leave your opinion in the TechTudo Forum.
How to make a slow motion video on iPhone iMovie