Have you seen the video of Barack Obama insulting Trump on Buzz Feed? Or the one of Tom Cruise with a lollipop, which is all the rage on TikTok? Well, they are not true, or at least transformed. I will explain to you as simply as possible the technologies involved and their challenges.
Want your idol to say he admires you?
This is now possible with deepfakes. However, what is the secret of this new technology that cannot put you into words? These days, deepfake videos can be found everywhere. It is easy to find deepfakes on social media. Originally, only porn sites used this process to replace the extra face with the face of a known personality. The result is astonishing–proof of the attractiveness that this practice has generated on these sites. Deepfakes are widespread and allow, taking into account modern achievements, to achieve, for example, parodies of celebrities, films or fake speeches of politicians – for example, in this video, we see the speech of a person imitating Trump on the left, and on the right, putting his voice in the face of the real president.
Concretely, what is the deepfake?
Definition of this practice
Deepfake is a contraction of deep learning and fake.
Deep learning brings together all the digital technologies that set up a system of algorithms in order to develop “amplified intelligence”.
Anglicism “fake”, on the other hand, means “false, wrong”. Thus, Deepfake is a “deep learning technology capable of synthesizing realistic fake videos of people making statements they never made.”
Deepfake Tom Cruise, example of faked videos
Take the example of the fake Tom Cruise videos. The DeepTomCruise account posted in February 2021 a series of videos showing the actor crouching and falling to the ground, playing golf or feasting on a lollipop. The whole planet is on fire, what interest would the actor have in showing himself like this?
In fact, it was a YouTuber, Chris Umé, who created a deepfake by replacing the face of an ‘impersonator’ (Miles Fisher) with that of Tom Cruise. Nobody noticed anything. This special effects specialist had to denounce himself to discover the deception. The idea behind this illusion is to alert the public to the potential dangers of Deepfake. Chris Umé also stresses the importance of supervising this technology, which can be used for malicious purposes.
Here, the deepfake developed by Chris Umé allowed two levels of transformation, which are as follows:
The encrustation of the face
The first is the encrustation of the face. By taking a database made up of thousands of photos of Tom Cruise, the deep learning system made it possible to record a maximum of his facial expression and to make a synthesis in order to make the permutation between the face as faithful as possible. By Tom Cruise and that of the actor (Miles Fisher).
The transformation of the voice
Then, a software allows a transformation of the voice. He recovers the tone of Tom Cruise’s voice by combining it with the movement of his lips and then superimposes it on Miles Fisher’s face.
Through these two techniques, the video rendering is extremely realistic and allows limiting as much as possible the inconsistencies that one might notice in an image.
What is the point of such a practice?
We could finally associate the deepfake with a copy of after effect or other video tracking software. However, unlike the latter, its interest can be summed up in two points: its ease of access and its use within everyone’s reach.
Ease of access
If you search well, you can easily find a number of free, unlicensed, unregulated deepfake tools. In addition, amateurs with only basic computer skills can operate them. It is very easy to find software from Deepfakes. There are even portable applications, which allow, perhaps in a less perfect way than software, to make fake videos. Take the example of FaceApp, which allows anyone, without special skills, to apply a filter to age them.
In 2019, a scandal erupted: the Russian company operating FaceApp, was suspected of recovering users’ personal data for its own file, a practice that does not comply with the European Personal Data Regulation (GDPR).
Face swap is another common example of deepfake. It involves swapping your head with that of another person in one or more photos or a video. The Reface application, released in June 2021, makes it possible, for example, to use this face swaping technique by integrating the face of a known person on our head. In addition, the result is stunning!
Ease of understanding and use
Software that uses a Deepfake is very easy to understand. A very widespread technique is that of “Lip-Sync Deepfake”. This process allows a voice to be introduced into a video without realizing it. There are several simple steps to do this: First, you need a basic video. You must then make a vocal recording of the speech you want to implant, as well as the movement of the lips during this speech. These two data will be transferred as a neural network model that will synchronize the new speech to the original video.
This technique therefore requires research into “deepfakes” software, which, when you have chosen your video, automatically synchronizes all the data transmitted and modifies the voice in order to create the false speech very realistically.
The dangers of Deepfakes
You get it; Deepfakes are ultra-realistic and accessible to everyone. Then comes to mind the question of the hijacking of the videos for defamatory or identity theft purposes.
Using Deepfake transforms the content of any video. Handling becomes easier and easier at a time when social networks are ubiquitous and controls difficult to put in place. Which raises the question of the veracity of what we watch and how we use the internet. It becomes important to be careful with all the information and research found on the web. Disinformation takes on its full meaning when one becomes aware of the dangers it entails.
Can we still trust the images, what we hear and even what we see?
What follow-up to this practice?
The research world is trying to control the Deepfake process with different applications, with detection software being one of the most viable solutions today.
Detection software
What does such a technological advance portend? How to control fakes news? It is clear that the bewildering ease with which it is possible for anyone to invade someone’s privacy or steal his or her identity can be frightening. It is therefore necessary to control this practice.
The Deepfake Detection Challenge
Measures have already been put in place such as the DeepFake Detection Challenge (DFDC). Facebook leads this initiative straight from the United States. She suggests that participants try to find a program capable of detecting fake videos from real ones. So far, the most efficient model has an accuracy rate of 82%.
The Checknew
Other initiatives are emerging to prevent fake news. Checknew is a fact-checking column from France. The newspaper Liberation develops it. It offers content that has been checked beforehand in order to be sure of the veracity of the words and images contained in the article. This section uses, among other things, the Assembler software, developed by Google, which detects retouching carried out on an image to find out whether it is real or faked.
