n May of 2020, I sat down with virtual influencer Aliona Pole for an exclusive interview. During our conversation, we discussed a wide array of topics: how Aliona believes fashion shouldn't harm the environment, how consumption should always be rational, how humans should practice “digital hygiene” online, and much more. Aliona is a model and a digital fashion designer, directly practicing what she preaches.
Today, I take a look behind the virtual curtain of Aliona Pole to speak with one of her founders, Valery Sharipov, on the subject of deepfake technology and identify the role this tech plays in the virtual influencer industry.
In our increasingly synthetic world, virtual humans are gaining stage presence. AI personalities are emerging as talking heads and blog authors, advertisers employ virtual influencers to promote their message, and music artists host record-breaking concerts in Fortnite without the involvement of the actual artist.
In this synthetic, digital landscape, deepfake technology gains increasing prevalence in the technical arsenal of marketing teams, production studios, and music labels alike, along with game engine technology and generative neural networks in lockstep.
The future will be synthesized.
While virtual influencers made specifically using deepfake technology are accessible, they are limited in their quality and portability. Most virtual influencers are 3D characters built using game engines (CGI), while deepfaked virtual influencers are created by applying a face atop a human stand-in (neural networks).
Entirely deepfaked virtual influencers arguably have their pros, and I am excited to consult Valery on the subject, who also founded a platform for automatic deepfake creation using neural networks. I intend to learn about how he pairs his tech with his virtual influencer storyline out of Russia, Aliona Pole.
In a world facing increasing digital distrust and information scrutiny, I hope to discover: Why does the world need deepfakes? What benefits do deepfakes bring to social media, if any? Why is deepfake technology spreading so rapidly?
Let’s dive in, face first.
Christopher: Hi Valery, nice to speak with you! To start this off, I’d like to understand what potential opportunities deepfake technology enable for content creators?
Valery: Content creators are the new rock stars. The concentration of advertising in digital modes is so high that the average effectiveness of advertising campaigns has dropped dramatically. It is no longer enough to just show your creativity—you need to present the brand in the right context, and all of this is against the background of video as a dominant format.
As the role of video in advertising continues to develop, more and more video creators are emerging with a growing demand from a dynamic audience. Fortunately, new instruments are also emerging: deepfake is becoming one of the new ways of self-expression on video.
So, what caused the sudden popularity of deepfake technology even though it’s technically been around for quite some time?
Well, it was originally called face swapping. That is, until one internet user with the pseudonym “deepfakes” gained popularity on Reddit in 2017, thanks to the porn videos he published. He, to put it mildly, illegally substituted the faces of famous personalities. The videos quickly spread across the net, hence the name familiar to everyone.
Last year, deepfakes started to emerge from the gray zone of pornography and YouTube humor, instead becoming a solution to pressing problems. Some of these challenges include:
- How to create videos with those who cannot be brought to the set of a shoot (thanks to the COVID-19 lockdown)
- How to change the age of a character by decades (tell the creators of Mandalorian about it)
- How to revive heroes from the past (see Disney’s developments in the space)
Deepfaking has a number of benefits not only for real humans, but also for digital people.
Interesting, how do you think about “virtual” vs “deepfake” humans?
Virtual human content built on computer graphics has premiere quality, beauty, and showiness, but also takes time, money, and the manual labor of a human designer.
This format is not necessarily suitable for those who want to create a lot of content for social media, for example. A deepfake human is one made using a neural mask which avails the benefits of a simpler, cheaper, and faster implementation, while opening up interesting possibilities.
What role will deepfaked people play in the future of the media?
With the help of a virtual human, you are completely protected from risks associated with dealing with humans. Here are some of the benefits, just to name a few:
- A digital person will not put your reputation in danger
- A digital person will not get sick
- A digital person will not go to a competitor
- A digital person can appear anywhere
With the help of a neural mask, you can now create content on an industrial scale, and anyone can be the “actor” who wears the mask of your virtual star. Plus, deepfakes open up new content possibilities:
- Create content when celebrities are physically unavailable—just create a neural mask of this person and record as much content as you want.
- Enjoy additional income and time savings—it's a great opportunity for famous personalities to earn more and save time by creating a deepfake and transferring the rights to an agency for advertising purposes.
- Access many unique, cheap faces—with deepfakes, brands can now quickly and cheaply create video content with a large selection of unique faces through image generation.
At a top level, how does deepfake technology work?
There are many ways to replace faces, but deepfaking involves the use of an algorithm based on generative adversarial neural networks. Using machine learning, the algorithm transforms your supplied media (a virtual or real human face) and applies it to a human video where the synthetic face should appear. The result is your so-called “deepfake”.
In this process, training a neural network model requires more resources towards the end goal of creating a quality deepfake. The longer the neural network is trained, and the better your source code, and, ultimately, the better the deepfake will be.
What tools are available for deepfake creation?
There are open source and free neural network applications with deepfake technology that can be used to replace faces and create videos with the right person in the "lead role". The most popular open source libraries are faceswap and DeepFaceLab.
Unfortunately, most deepfake projects are currently handmade or marked up for each video or entertainment application, and they prevent you from deepfaking anyone you’d like—limiting your options to a list of low resolution, predetermined faces, often of celebrities.
How do you predict deepfake technology will evolve in the coming years?
I think more and more services will appear over time. In addition to the obvious things like improving quality, improving speed, and lowering prices, it's easy to predict other aspects. For example, generating an appearance according to specified parameters using neural networks, or using these images when creating deepfake videos.
Plus, we will see the generation of faces, voices, as well as texts in the style of characters. I say this with confidence because we are already working on all of this, and I am sure we are not alone in the industry.
As for digital content consumption, what impact do you think deepfake technology will have?
I am more and more confident that deepfake content will appear more regularly, especially with the reduction in speed & cost, a decrease in reputation risks due to the absence of a person, and new mechanics for creators.
Think, too, what use cases might this draw out? It's anything—there are many options, and the limit is only your fantasy. For example, I've always been inspired by the idea of playing multiple roles in my own film—I think it's cool!
A few cases we will likely see in reality soon enough:
- Creating a digital twin of a real celebrity, which is difficult to regularly involve in real filming
- Using the image of a historical or fictional person in advertising
- Creating a fictional character and content with him (this can be a solution for those who value anonymity)
- Creating comedic content and parody videos
- Above all, digital heroes and neural networks are driving down the cost of creating special effects content
What are some of your favorite deepfakes?
A couple of my favorites include Steven Wilson's DeepFake music video SELF and the Sassy Justice comedy show from the creators of South Park.
There was also a cool commercial in which David Beckham talks about the dangers of malaria in 9 languages. Another awesome case is deepfake TikTok profile of fake Tom Cruise—fantastic job. By the way, our Aliona Pole also has a page on TikTok, made with a neural mask.
And finally—I have a special place in my heart for when the Salvador Dali Museum in Florida "revived" the artist. As a result, viewers could talk to Dali, listen to his stories and even take selfies together. It’s also amazing because, during his lifetime, Dali was sure that he would not die and dreamed of becoming immortal. Now, his dream has come true.
So, I wish all of us will dream big and make these dreams come true.
Thank you, Valery, for the fascinating look at deepfake technology!
Sure thing. It was a pleasure for me and I hope this interview will inspire creators to generate even more amazing content with deepfake technology.