Deepfake: Understand what it is, how to identify it, and what risks it offers. ¬†¬†¬†¬†¬†¬†¬†¬†¬†¬†¬†¬†¬†¬†¬†¬†¬†¬†¬†¬†¬†¬†¬†¬†¬†¬†¬†¬†Deepfake is the name given to the technique that allows to recreate the human appearance or voice with the aid of artificial intelligence. With technology, it is possible to simulate people talking or acting …

Deepfake It is the technical given name that allows to recreate the human appearance or voice with the aid of artificial intelligence. With technology, it is possible to simulate people speaking or acting in ways that they would never do.

The word would be a junction of terms deep learning (a branch of machine learning or machine learning in Portuguese) and fake (false). The concept would have emerged in 2017, in reference to a Reddit user with the nickname of deepfakes.

He would have used the technology of deep learning to overlay celebrity faces with those of people in pornographic films. Since then, the method has become increasingly realistic and affordable.

AppGeek has prepared a guide for you to better understand what and how the web-based editing feature works. Also check out what are the risks offered and how to protect yourself.

How does the deepfake?

DeepfakeActress Michelle Adams has face replaced by actor Nicolas Cage (Reproduob / Wikipedia)

To create a video deepfake two competing artificial intelligence algorithms, called generator and discriminator. According to Whatis, the function of the first is to create fake videos and the second to identify them as true or not.

When the discriminator finds a false content, the generator learns what not to do. And so you will create the next video in an improved way. When acting together, generator and discriminator generate the call generative adversarial network (GAN), or generative opposition network, in free translation.

How to create a video deepfake?

To make a movie of the genre you need to have a source video, preferably as simple as possible. Therefore, political speeches are often used, in which the neutral background is basically made of head movements.

So I need to combine two datasets. If the video character turns his head to the left and opens his mouth, it will be necessary to have images of who will replace him in the same positions. This may require a vast image bank.

Then it's up to the job to leave the program based on artificial intelligence. After several trials and errors and successes, the generative adversarial network will incorporate one face to the other as accurately as possible.

The video above is one of the most famous of deepfake In it, former United States President Barack Obama has movements and voice overlaid by those of comedian Jordan Peele.

What are the risks of deepfake?

Although it seems harmless, the deepfake has been controversial since its inception in 2017 at Reddit. The technique was often used to superimpose the faces of actresses on people in pornographic videos.

As it develops, the technique becomes more realistic and the contents end up viralizing as if they were real. This is one of the great dangers of the resource: the spread of false content as authentic.

In 2018, Facebook found itself involved in a scandal following the leakage of user data by a third party application company. Shortly after, a video of Mark Zuckerberg went viral and worried a lot of people.

In the short movie, Facebook's CEO insinuates that he controls the future by having the data domain. Content was nothing more than product deepfake created by Bill Poster for the company Specter.

Another problem predicted is the loss of credibility of any type of material posted on the internet. Technology makes it increasingly complicated to identify and filter what kind of news is actually or simply created by the computer. What your eyes and ears come and hear may be a software product.

With the popularization of programs that allow lay people to create such videos, it is possible that ordinary people are also increasingly targeted by this kind of montage.

Results can range from simulating participation in adult movies to creating false evidence against one.

As a recent phenomenon, governments do not yet have specific laws to combat crimes related to deepfake. In the United Kingdom, the producer of such content may be indicted for harassment.

In the United States, charges can range from identity theft, cyber stalking, and revenge pornography. In Brazil, there is still no specific regulation on the subject.

How to identify one deepfake?

Despite the evolution of the technique, most of the deepfakes amateurs can still be identified with the naked eye. In such cases, it is possible to identify unnaturalness in the blink of an eye, lip movement or the presence of shadows in the wrong places.

However, concern about the real future. In the United States, the Advance Defense Research Projects Agency (DARPA) would already be investing in ways to better identify videos.

For now, it is best to always be alert to what you receive and be aware of sharing sources.

What the ZAO?

Announced in September 2019 on the Chinese App Store, the app ZAO became news all over the world. The program went viral by leading smartphones to create deepfakes from a photo in just 8 seconds.

The result was perfect and was shared on Twitter by user @AllanXia. The face of actor Leonardo DiCaprio has been replaced by his in several scenes and the expressiveness impresses.

In case you haven't heard, #ZAO is a Chinese app which completely blew up since Friday. Best application of 'Deepfake'-style AI facial replacement I've ever seen. Here's an example of me as DiCaprio (generated in 8 secs from that one photo in the thumbnail) pic.twitter/1RpnJJ3wgT

Allan Xia (@AllanXia) September 1, 2019

The app launch has reopened the technology and privacy debate. Firstly, according to the terms of use of the program, by using it, it would be giving the company the right to use its data and face for free and irrevocable commercial use.

After many complaints, the company responsible changed the term. It now states that the content will only be used by developers for application and technology enhancement purposes.

Another worrying issue is that the app makes it even simpler to create overlays with anyone's images. And with a growing realism. This could make it easier for victims of fake videos to appear to be real.

Other apps by deepfake

But this is not the first time that deepfake viralize. The first one was the Fakeapp, released in January 2018 for desktop. It made it easy to create and share video with replaced faces.

The main difficulty was to gather a large amount of visual material from the person to be inserted, so that the result was satisfactory. Currently, we can no longer find the program available.


Already in 2019, the software was released Deepnude, for Windows and Linux. The app was able to use the photo of a dressed woman and leave her naked in a few clicks.

The fake image was accompanied by a watermark, but nothing that little image editing knowledge would allow to remove. In June of the same year, the company announced that the program would be discontinued.

Interested in apps that allow you to change your look? So meet Snapchat. The famous app lets you change faces with friends and apply face filters to photos and videos in real time.

Also check out 7 apps to turn your photos into drawings easily and 8 super fun apps to change your voice.