Deepfake videos are fake productions that use machine learning and artificial intelligence to change people's faces and put them in situations they have never experienced in real life. The realism of the technique, which has begun to be used to produce celebrity mounts on pornographic videos and has already helped forge speeches by influential politicians, worries experts and entities such as the Pentagon and the United States Congress. In the following list, you will find seven curiosities about deepfake polymers, considered the evolution of fake news.
READ: Fake porn videos use celebrities and generate controversy; Look
Deepfake video with Mark Zuckerberg causes controversy in the United States Photo: Playback / Instagram
Want to buy a cell phone, TV and other discounted products? Meet the Compare dnetc
The term deepfake first appeared in December 2017, when a Reddit forum user by that name began to transpose famous faces into pornographic videos. In the same month, Vice Magazine published on its website an article denouncing a manipulated video that showed the actress Gal Gadot, star of the movie Wonder Woman, having sex with a family member. This fake was done by deepfake, who also used machine learning software to apply Taylor Swift and Scarlett Johansson's faces to existing clips.
Actress Gal Gadot's face was applied to another woman's body in fake pornographic video Photo: Reproduction / Vice
It was noted that the videos were made by one person who was able to easily and quickly create convincing and high quality counterfeits. With this, the expression deepfake was soon used to indicate a variety of edited machine-learning videos and other artificial intelligence capabilities.
2. Companies offer the service
In 2018, a pornographic company took advantage of the popularization of deepfakes to offer a new service to its customers. Naughty America's proposal was to help them realize a sexual fantasy by putting their faces on that of porn actors and vice versa. Consumers were given a script that directed them to film themselves with certain expressions to ensure that the end product was realistic.
DeepNude only worked with pictures of women; app was discontinued after controversy Photo: Divulgao / DeepNude
Another case of deepfake monetization involved the DeepNude app, which created photos of naked women out of artificial intelligence. With a few clicks, the app swapped sweaters for breasts and shut for vulvas, producing quite realistic nudes. In June this year, after negative repercussions in the international press, the application was discontinued. Still, alternative versions of DeepNude continue to circulate on the Internet.
3. Facebook has already refused to take down deepfakes
Last month Mark Zuckerberg appeared on an Instagram video claiming to control data stolen from billions of people. The fake production, which bore the CBSN news channel logo, used the deepfake technique and was created by two artists in conjunction with the Canny advertising agency. Although CBSN claimed to use the trademark, Facebook refused to remove the video, stating that it would only remove the material if the falsehood was attested to by third party checkers.
4. Deepfakes in the sights of the US Congress
The possible impacts of deepfakes on the US presidential elections of 2020 led the US Congress to take action. In December 2018, the first federal technology bill was created. The Malicious Deepfakes Ban Project (Malicious Deep Fake Prohibition, makes it federal crime to create or distribute deepfakes where it facilitates illegal conduct.
The Deepfakes Accountability Act, adopted in June this year, would require mandatory watermarks and clear labels on all deepfakes – a rule that is likely to be ignored by anyone who is interested in turning fake video into a political weapon. The Congress is also studying fighting deepfakes by regulating social networks.
5. Deepfakes can be created by lay users.
The FakeApp software release in January 2018 made the creation of deepfakes video accessible to regular users. You can find tutorials that teach you how to use the tool on Internet forums.
In fake video, Obama's moves and lines are controlled by actor Jordan Peele Photo: Playback / BuzzFeed
A few months later, in April, BuzzFeed published a video in which former US President Barack Obama spoke words that were not his to warn of the dangers posed by technology. The video, which in the early stages claimed to be a scam, was made by a single person using FakeApp. The program took 56 hours to process the original materials.
6. Deepfake Technology Cannot Create 3D Faces
Real videos show people moving their faces in three dimensions, but deepfake algorithms are not yet able to make faces in 3D. What they do is generate a two-dimensional image of the face and then try to rotate it, resize it and distort it to fit the direction the person should be looking at. This explains, for example, why the faces of deepfake clip characters often have some misaligned features and contours.
7. can identify deepfakes
Although they are sophisticated, deepfakes videos are not perfect and can be identified by looking at some aspects. First and most glaring that, in general, deepfake characters flash much less than a normal person. This is because blinks are often not included in the training algorithm data set.
Scientists work to combat deepfakes Photo: Reproduo / Discover Magazine
Another point, already mentioned, is the alignment of facial features, which generally has a wax texture in the less convincing fake productions. Also, it is noteworthy that deepfake videos consist only of the exchange of faces, ie if the body of the person involved appears to be much thinner, heavier, taller or shorter than in real life, there is a great chance of being fake .
In this sense, among the indications that false production is also the absence of sound or a lip syncing artificial, where the words spoken do not correspond correctly to the movement of the lips, or the lips move in a strange way.
Lastly, it is important to note that deepfake videos are usually no longer than one minute long, as processing a few seconds takes hours and demands a lot from the machine that operates the software.