We all have seen Mark Zuckerberg screaming about having total control of million people or our favorite movie actor talking garbage. So, we have witnessed deepfakes. In this century, it is a time of Photoshop. If you want to star yourself in your favorite movie or play any sport with a famous athlete, it’s time for you to make deepfakes. So, What is Deepfake? To know more, read the whole article.
What is Deepfake? How Does It Work?
The word ‘deepfake’ was derived from deep learning, which is a form of Artificial Intelligence (AI). Its algorithm teaches it to solve a large set of problems. It can swap faces of the video in the digital media, and they look natural.
Deepfake technology can put someone in a spot where he never participated. Before it years to make a perfect video. But with the increasing machine learning system, it has become more accessible. On social media, you must have seen these kinds of things.
This technology is with us for many years. But it didn’t make any significant difference. Until 2017, a post on Reddit showed us how an AI could swap faces easily. And that was called deepfake. Nowadays, there are many processes to swap faces. Some of them need AI; also, we can make deepfakes without AI.
How is Deepfake Made? What Type of Technology Do I Need?
Machine learning is the main component to make deepfake. First, the programmer has to teach the machine about the persons whose faces he wants to swap. The creator has to provide hundreds of photos and videos of the persons in an AI algorithm called Encoder. The machine will examine the persons’ movement from different angles and lighting and find the similarities between them. Then it will reduce their similarities; also, it will compress the image.
Then the second algorithm, ‘Decoder,’ comes in to recover the compressed images. You need two decoders for two faces. Here, you need to put the face in the wrong decoder. That means A’s face will be given to B’s decoder and B’s face to A’s. Thus the decoders will swap the faces. To make it realistic, you will have to do it frame by frame.
Sometimes the creator has to add more information manually to avoid unnecessary glitches in the video. The process is not very straightforward. There are many other processes of making deepfakes. As technology is evolving, many more will come in the future.
It’s not easy to make deepfakes on standard computers. It takes a high-end desktop with a massive graphics card and better computing time with the cloud to do so. It takes lesser time to do.
Also, only the machine can not do it. The creator must have a particular skill to do the touch-ups. There are many software, even mobile apps, which can make deepfakes for you. But to make it perfect, you need to have a certain amount of skill sets.
How to Identify Deepfakes?
It gets harder to identify deepfakes as technology is evolving day by day. Most of the machines can not detect eye movement. So in most of the deepfake videos, the person does not blink or blinks unnaturally. Also, the eyes sometimes react independently of each other. But the videos are in low resolution to hide these issues.
Most of the deepfakes available online are made by amateurs. The lip-syncing is very bad, or the skin tone may not match. They can hardly fool you.
The process makes evident artifacts around the face of the person in the video. Sometimes it will blur or flicker when the angle of the face changes or will miss the details of the hair.
The lighting may not match the video. The deepfake algorithm follows the model lighting to make a deepfake. That will make a poor match with original lighting.
If a video is faked and the audio is not manipulated carefully, the audio will not match the original one.
But, with the advancement of technology, some deepfakes are perfectly made, and we cannot identify them.
Are All Deepfakes Videos?
No. Deepfake technology can make a compelling but fictional photo from almost nothing. There are some examples we can experience in social media. Also, you can make deepfake audios. You can create voice skins or voice closes of celebrities to make them tell whatever you want.
How are Deepfakes Used?
Most of the deepfakes are pornographic. In September 2019, an AI firm discovered 15,000 deepfake videos online. It nearly doubled after nine-month. There, 96% were pornographic, of which 99% of them were celebrities who swapped faces with pornstars. With modern technology, as amateurs can make deepfakes, it will cross the celebrities’ boundary and become a weapon of revenge.
A vast majority of deepfakes are of politicians and celebrities. A large number of photos are needed to make deepfakes. So the celebrities and politicians are more vulnerable.
Also, it is a threat for corporate employees. You can easily deepfake audio to confirm a file, which is not established initially. It will make a big scam.
Lastly, if you can make pornographic of celebrities, you can make them of athletes, politicians or even general people like us.
Can We Use Deepfakes For Our Good?
Not at all. Deepfakes has a fair amount of use for entertainment purposes. A deepfake of a late painter in the Dali museum will introduce you to his art and take selfies with you.
Researchers are working to understand this technology more. With the help of reliable and accessible technology, they are using it to edit videos and images. Deepfakes will be used in the film industry. It can slowly replace make and costumes.
The film industry sent photographers or videographers to shoot the original places to use them in their movies, which was risky, time-consuming, and expensive. But with the introduction of deepfakes, it has become straightforward.
Also, with deepfakes, the film industry can make young age of old actors take their young age shoots. Also, they can use the face of an actor who is not even alive. It is handy for them. For example, if an actor dies before finishing the shooting of a film, they can use deep fakes to finish it.
Gaming and fashion industries may benefit from this. They can make games or costumes with their desired celebrities or athletes.
Deepfakes can restore people’s voices which they lost in some accident.
Lastly, it can be used to make movies dubbed into different languages to make them global.
What Is The Future of Deepfakes?
There have been many talks about the use of deepfakes. It threatens mainly the politicians. Strict rules should be applied in the service of deepfakes. It will slow down or even close some researches on this AI. Many websites have rules against deepfakes. They remove all kinds of no-consensual deepfakes.
On the other hand, the unrestricted use of this technology can hamper our daily lives. You can easily digitally recreate any historical figure or late celebrity to use them as you want. So, it is a matter of time that strict rules and regulations for the use of deepfakes will come.
As technology is getting better eventually, it will be tough for us to detect deepfakes.
How Can We Prevent Deepfakes?
Are deepfakes illegal? There is no straightforward answer as many use it for good purposes. But some states in the USA have laws to prevent deepfakes. Outside the USA, China and North Korea have taken actions against it.
The pornographic sites are taking steps to reduce deepfakes, as it is mainly used in it. But it isn’t easy nowadays to detect deepfakes. There is no chance you can make a deepfake detector. The creator will use it to make their deepfakes better.
Tech companies are working to make deepfake detectors. Artificial Intelligence is the answer also. The footage of the celebrities will train the software. It will find the origin of the video and if any manipulation is used. It will notify as soon as it detects any deepfake.
Things become blurry when it comes to politicians. Lawmakers are taking necessary actions to solve this problem.
As we have learned so far, deep fakes can be used in many good ways. But till now, it is used chiefly to harass or defames people. So, we should be concerned about what we should trust on the internet and double-check if anything seems suspicious.
Subscribe to our newsletter
& plug into
the world of technology