The technology behind deepfakes is more sophisticated than ever before. Unfortunately, that also creates more risk in terms of the potential damage these videos can cause.
Learn more about how deepfake videos are made and why many are concerned about their potential effects.
What is a deepfake video?
A deepfake is a type of synthetic media that has been manipulated to feature someone else’s likeness.
Using artificial intelligence, the face of a person in a video or image can be swapped or manipulated. Even the audio can be synthesized to sound just like the person whose face is being used.
The impressive advances in technology in recent years has made it very difficult to detect when an image or video is a deepfake.
For an overview on deepfakes, see this in-depth explainer.
How are deepfake videos created?
The term deepfake actually comes from a combination of “deep learning” and “fake.”
Deep learning is a type of machine learning that makes deepfake videos possible.
Using artificial intelligence, large sets of data can be processed and analyzed to swap faces and audio in order to create a shockingly realistic fake video.
This can be accomplished through various learning algorithms. Initially, neural networks analyze a series of video clips of the person who will be face-swapped into a target video.
The artificial intelligence algorithm uses that data to learn the person’s face and movements and map it onto the target video.
Once that step is complete, a different algorithm uses generative adversarial networks (GANs) to detect any flaws that could flag the video as a fake.
These networks have already studied large amounts of data to learn how to fix those issues so the finished product looks more natural.
This is like the fine-tuning step for the deepfake that catches any issues and makes sure that the video looks as authentic as possible.
Other artificial intelligence algorithms can be used to create images or videos of people who don’t exist at all, which is another type of deepfake that does not involve using the likeness or voice of a real person.
What software is used for deepfakes?
There are different types of software used to create deepfakes. Some open-source software options can be found on GitHub, such as DeepFaceLab.
These kinds of software are fairly advanced, so users typically need some tech training in order to follow along with available tutorials and use them effectively.
Additionally, open-source software options for creating deepfakes generally require a considerable level of CPU and GPU in order to successfully create a high-quality video.
People who have the expertise and computing power, however, can potentially create very convincing deepfakes in this way.
Are there apps that can make deepfakes?
In addition to advanced open-source software, there are also a number of applications that allow people to make their own deepfakes.
Depending on the type of app used, the results may not be as realistic as those produced by high-tech software like DeepFaceLab.
Some deepfake apps are capable of producing quite convincing results, like FakeApp. This desktop application can analyze thousands of images and videos to create highly realistic deepfake videos.
For those with less tech expertise, phone apps can be a more accessible alternative for creating deepfakes—although they often come with limitations.
An app called Reface, available for both iOS and Android phones, superimposes faces into videos and GIFs. In China, a similar app called Zao is available for Apple phones.
These apps only have a limited selection of video clips available (mostly those featuring famous actors, singers, works of art, or video game characters).
Additionally, the images mapped onto videos or GIFs must come from the phone’s front-facing camera. The idea is that users can only create deepfakes using selfies rather than images of other individuals.
There are also a number of free face swapping mobile apps available for iPhone and Android phones that are designed with beginners in mind. Face swap filters are also available on social media platforms like Instagram and TikTok.
The quality is not convincing with these options, so the videos don’t pose the same types of risks as highly realistic deepfakes developed with sophisticated technology.
Despite these apps being designed with entertainment purposes in mind, they also offer insight into how unsettling it can be to see one’s face superimposed on another person’s body.
How are deepfake voices made?
To make synthetic voices, programs need to be able to analyze voice samples rather than video clips.
These versions of deepfake software apply machine learning and artificial intelligence, but with a focus only on audio.
The voice samples uploaded to the software are eventually merged, creating a fake voice that mimics the person in the original recordings.
Once that happens, the person using the software can make the voice say just about anything they want it to in real time using a text-to-speech program.
Deepfake technology has improved to the point where very convincing synthesized voices can be produced.
What kind of damage can deepfakes cause?
As the technology continues to improve, deepfakes are becoming more and more realistic—which also means they are more capable of deceiving viewers or listeners.
This can create a number of serious problems that can lead to personal, professional, financial, and even political damage.
The following are just some of the issues that can potentially arise from the use of highly convincing deepfakes.
The first deepfakes to gain a significant amount of traffic on the internet were pornographic in nature. Most featured the faces of famous Hollywood actresses mapped onto those of participants in erotic films.
This type of misrepresentation could be construed as defamation, especially if the videos are effective in deceiving their audience.
The videos could have a number of negative effects for the person whose face has been used, including loss of employment opportunities, damage to their personal reputation, and lasting psychological effects.
The creation of disinformation content that is intended to mislead the public is a major concern with deepfakes today.
Many are worried that deepfake videos or audio could be used to spread fake news or bolster conspiracy theories.
Sophisticated video editing could potentially be used to develop deepfakes featuring prominent world leaders. If one of those videos go viral before it’s detected, there are concerns that it might create chaos and unrest.
Deepfakes also have the potential to play a role in fraud for personal or financial gain.
For example, deepfake audio has been used to make phone calls requesting financial transfers. Because the calls mimic the voice of a person in authority, employees follow instructions and unknowingly send money to criminals.
Similar types of fraud can occur by using deepfakes to access bank accounts and other private information protected by voice recognition technology.
Mistrust of media
Video and audio recordings have long been thought of as hard evidence of an event. But with deepfakes, these media sources can be called into question.
As people become more aware of deepfakes, they may also become skeptical of the things they see or hear in these recordings.
Furthermore, the existence of deepfakes makes it possible for people to mistrust what they see or hear. Some may even be inclined to deny the veracity of a video or audio recording due to personal beliefs or political gain.
It’s important to recognize that, while not all deepfakes are created with the intent to cause harm, they can result in serious damage, including everything from personal financial losses to political upheaval.