What is deep fake and how it works?

August 11, 2020
Do you already have been amused by Donald Trump as Saul Goodman in Breaking Bad video? Or maybe do you have seen Jim Carrey as Jack Torrance in The Shinning movie or Robert Downey Jr. and Tom Holland in Back to the future? Have you terrified with Mark Zuckerberg declarations about Facebook privacy? Did you see Obama’s public service announcement already? All these videos are generated with something called deep fake. Well, in this article we will explain you what it is and how it works.
What is deep fake?
Generally speaking, deep fake are look alike multimedia generated by a computing device with a specific technology called deep learning. In fact, its name comes from “deep learning” and “fake”. Deep fake refers to any fake video, photo or audio manipulated or created by using deep learning, which is a machine learning method, a part of artificial intelligence.
In most cases, it is used to identify a person gestures in an existing source and swap faces or voices nuances by replace them with similarities in someone else.
How deep fake works?
Deep fakes leverage powerful artificial intelligence technologies to generate multimedia content that intends to mislead or to making seems something as it would be true. It is based on deep learning that allows to generate content faster and at lowest cost. It does not require a special computing device very different from we already have, but it needs a high computing processing in order to work.
Without further details and advanced knowledge, to be able to generate a deep fake we would have to train a neural network with a generative method working together with advanced computer graphics methods to combined someone or something onto another target. Most popular frameworks to do this use generative adversarial networks (GAN), but there are others models too. Is it not a simple work, it requires a lot of a person images or audio sources in order to feed a neural network to get the job nicely done; while greater effort and sources used, the better it would be the deep fake quality.
By example, to make a deep fake video we need a large subject video footage to train the neural network in order to find similarities with this and its target and reduces the noises that cause differences between them. That is why celebrities and politicians are principal protagonist of most videos, because there are an open wide sources of them over the internet.
Today however, we do not need so much knowledge or working hours to generate short deep fake videos with some quality thanks to several applications and sites already available that we will talking about next.
Deep fake apps
There are a plenty of apps available that do not have high or powerful requirements, and you can use it for academic purposes o just for fun to generate your own deep fakes:
FakeApp
FakeApp is an easy to use desktop application that allows to create and share deep fake videos swapping the user face on it.
FaceApp.
FaceApp is a mobile application to generate deep fake images and videos based on people faces transformation. By example, it can make the user seems younger or change their gender.
Face Swap Live
Face Swap Live is a mobile application that uses the camera in the device to let users swap faces with someone else in real time.
DeepFaceLab
DeepFaceLab is a software that provides a framework to create deep fakes, it is an open source application that does not need further understanding of deep learning.
Zao
Zao is a mobile application that can be used to superimpose user into celebrities faces and to modulate their voices, mostly in famous movies scenes.
Celebrity Face Morph
Celebrity Face Morph is a mobile application that permits user be morphed into a celebrity face.
Doublicat
Doublicat is another mobile application that let users swap their faces on to a gif or meme found in its library.
We can use all of these tools and more, even other capable of generate audio deep fakes, easily and without powerful device requirements, but be cautious and respectful to not violate privacy of anyone.
Deepfake concerns
Although deep fakes technologies are evolving and growing, the fact that they are more accessible to common people has raised too much concerns about how it will be use it. One of the main concerns about deep fake is how hard is to detect it, both for humans and through automated methods, this has risen so much controversies due to its misuse, especially in matters of public persons like politicians and actors, which there are a large footage bank available to build almost any multimedia material, even from scratch. Its malicious use can even generate national security problems by causing greater damages.
Is common to find celebrities pornographic videos, fake news, or personal revenge porn. Its proliferation is a matter that have worried even government and industry what has led them to discuss and take actions in order to limit or detect their use. Today, there are so many laws and terms conditions that aims to forbid people making and publish them.
Facebook, Twitter, even platforms like Tik Tok are taking actions to protect them against content manipulation that can be occurs in misinformation. Some of the politics are banning materials with non consensual content or flagged them as deep fakes.
So, talk about what is deep fake and how it works implies to take those things in consideration, in spite of faking content is not a new problem, but its potential could be so useful as harmful.
Let’s do a good use of deep fake if we need it, or can enjoy of the greatest things that are created with it today.
Have fun but don’t be fake!