8 Ideas on How Deepfakes Will Change Photography and Marketing in 2020
Deepfakes — the digitally created visuals that are so realistic that it is almost impossible to distinguish them from real photos or videos. It is not a new technological phenomenon, but a scary concept for many.
Cinema gave us the concept of a mockumentary (pseudo-documentary), as well as digital technologies for creating realistic “special effects” on screen. But only in recent years, pseudo-documentary images and videos without any special labeling, the synthetic nature of which cannot be recognized easily, have gone beyond the scope of art and become an instrument of socialization, marketing, political manipulation, cybercrime, and, possibly, a new creative method.
At Depositphotos, we keep abreast of trends that can affect the development of photography and marketing. This article will help you understand the nature of deepfakes and discover the potential of new technologies in the context of your creative projects.
The post-truth epoch is welcoming you! What deep learning fakes are
There is no point denying that you have already experienced deepfakes in a digital environment. Remember apps that generated photos of your “old” face or offered you to see yourself in the male or female version? And what about face filters, which initially allowed the exchange of faces during selfies and other manipulations with the faces of real people?
One of the earliest and straight-forward examples: the remastered image of Princess Leia from Star Wars: Rogue One, which was created by applying a model of the face of a young Carrie Fisher to the face of another actress who was playing the role. The technology was very simple. The creators simply cut out the image of Carrie Fisher’s head from the early Star Wars episodes and pasted it into a new video.
Another compelling use case for deepfakes is related to politics: Barack Obama videos, created by BuzzFeed and comedian Jordan Peele, as well as videos based on the AI model of the US ex-President, developed by the University of Washington.
What are deepfakes and what is their difference from pseudo-documentary, which is as old as photography itself? The term deepfakes comes from the name of the deep learning technology, which is a branch of ML (machine learning). This technology exploits neural net simulation, which learns and improves by data arrays processing.
The more data passed through a deep learning neural network, the more convincing the fake generated by the system will be.
Although one can create a deepfake voice (example: voice clones of a politician and ZAO application), a large number of deepfakes are photos and videos. In 2020, a large part of deepfakes is faking public figures, including politicians and stars, but there are also several cases where cybercriminals used deepfakes of CEOs to receive money.
How are deepfakes used today? We could endlessly discuss the horizons that the deep learning technology opens for fiction film production and education, but the real facts are disappointing. Recently, the Deeptrace company found over 30,000 deepfake videos online and around 96% of them turned out to be pornographic (with mapped faces from celebrities). The reason lies in the fact that modern deep learning technologies have become available to a wide range of users, including hunters for easy money.
Top Technologies Used for Deepfakes Generation
As mentioned, a deepfake is not just an image that realistically reflects an event that actually had never happened. Deep learning technologies as we know them were presented in 2017. Everything that existed up to this point (mainly in cinematography and ad production) was a synthesis of semi-automatic post-processing and good scenery.
An important feature of deepfakes is that they are created by an AI algorithm (a program that can train itself), and not a person.
⚙️ Encoder-based technologies
An algorithm called an encoder compares two images (these can be images of faces, bodies, or other objects) and finds similarities between them. For example, the first picture shows the person you want to simulate and the second picture shows the one who will play the role of the first person. The decoder reconstructs the first object with the expressions and orientation of the second one.
The basic encoder technology has become the prototype of many other algorithms that relate to facial and object recognition, which are now widely used in social networks and online stores.
Tools: FaceApp, Doublicat, FakeApp, DeepFaceLab (95% of deepfake videos are created with it!), Machine Tube
⚙️ Generative Adversarial Networks (GANs)
The idea behind GAN is that it allows laptops to not only classify images but also create them. GAN is a synthesis of two AI technologies: generator and discriminator. The generator creates an image by randomly selecting elements of different images from the internet. The result of the generator does not look like a real image of a face or another object, it is a literally random set of pixels.
The discriminator’s task is to edit this file so that it becomes similar to the images that its AI core processed (for example, celebrity photos from Instagram). Such a cycle can be repeated several times to achieve a better result, that is, to get an image of an object that is very similar to the real one, but is 100% synthetic. GAN can also generate music and voices.
GAN-based tools: HyperGAN, CycleGAN, RoboCoDraw, pg-GAN (Nvidia)
The two technologies described above can also be combined. For example, you can generate a realistic avatar with GAN and then turn it into an Instagram face filter and create real-time videos pretending to be somebody else with encounter technologies.
Deepfakes Impact on the Photography Industry
The development of digital technology is a natural process, so we recommend that you not ignore it and think about the benefits that deep learning technologies can bring for photography as an art form and business industry. Here are some inspirational ideas from Depositphotos:
#1 The birth of a creative method
Deepfake artists already exist. One of the pioneers in this area is called Hao Li. Bill Posters and Daniel Howe, who run an Instagram account on social topics (fake Mark Zuckerberg video is their most famous work) also call themselves artists. Look into Collider Videos and Gillian Wearing if you value highly artistic works. For now, we can see that deepfake’s ability to imitate reality makes it an ideal tool for criticizing society in a comic or sarcastic manner.
#2 Art exhibitions anywhere in the world
Deepfakes have made the art of photography more accessible. The technology allows you to recreate art objects anywhere in the world (not only in Paris and New York galleries), and also means a deeper interaction between visitors and digital art objects.
You can create a deepfake inside any gallery and simulate a personal exhibition there. In turn, visitors to the exhibition can feel like a part of the art objects using deepfaking apps, or interact with them. An example is the Dali Museum in Florida, where the artist himself meets guests and the smiling Mona Lisa from Samsung’s AI lab.
#3 The profession becomes more democratic
Not all photographers can afford top-class photo equipment or professional models for an art photoshoot. Digital technologies, such as deep learning, require only a powerful computer or even a mobile phone. Thus, any photographer can take several pictures (not necessarily high quality), and then synthesize them into one, and also use virtual characters and objects to create the necessary setting.
#4 Take photography to the next level
Novice photographers will now be able to learn faster and spend less money on their education, as deep learning technology can not only create photo fakes but also point out compositional mistakes to a photographer, give them advice and gradually train them. Such computer programs based on AI, which operates with a huge legacy of world photography, already exist (check our Skylum tools).
Another opportunity for photographers is to sell mockups for deepfake photo collections with a client’s face and body.
#5 Photographers are needed to create content for AI processing
We can expect that soon there will be a demand for those who produce content for AI training. Remember that deep learning technologies need images to improve themselves. The more images, the better. At the same time, specialized AI-based products need certain images (for automated diagnosis of diseases, images of healthy and sick people are needed, and celebrity reports are required to generate celebrity deepfake on Instagram). Perhaps soon such photo databases will cost a lot of money!
How advertising will change in if technologies continue to evolve
For some areas such as litigation, state defense, or journalism, deepfakes pose serious threats, and others (like marketing) benefit from deep learning technologies a lot. AI-powered images help brands provide customers with a more personalized user experience, help them create a future purchase or service feel, and establish friendships between the brand and customers by imitating real-life interactions with company staff. And there are more other options.
#1 Sense of reality for better PR and advertising
There is a bright side to deepfakes too. This technology is capable of generating images that stimulate loyalty to the brand when it is impossible to take photographs. More creative PR specialists are exploiting the technology in order to share brand messages better. Example: the social campaign called Malaria Must Die where deepfake David Beckham was speaking about Malaria in different languages. The voices belonged to doctors and malaria survivors.
#2 Imitation of product ownership
Marketers know that the likelihood of a sale increases if users have already felt that they own the product. Deep learning technology will help simulate this situation. It can show the potential buyer of cosmetics her face with a future makeup look, or impress a car showroom customer with a deepfake photo of him behind the wheel.
#3 Progressive (and cheap!) localization
It’s not necessary to shoot ads for each local market, using CGI models and the right local decorations to help customers recognize themselves in advertising. Brands can create only one advertising video, and then remove and add elements to it to gain positive impressions of customers throughout the world.
This approach also makes it possible to save costs on production, as deep learning technologies give us more possibilities to challenge the limitations of photography.
Final thoughts
It seems too early for panic over the widespread adoption of deepfakes. No matter how scary it is that an ordinary person can not distinguish a high-quality deepfake from a real documentary, the technologies that made such images possible take the photography and marketing industry to a new level of quality and possibilities.
In addition to the emergence of a new art field and a whole galaxy of influential artists, the development of deep learning technologies made the profession of a photographer more open for newcomers, and also reduced the cost of photo production. As for marketing, there is room for optimism too. Deepfakes allow brands to win the hearts of potential customers and serve people from a distance, helping them to choose the product they like best.
As a high-tech company, whose work is also based on AI, Depositphotos closely monitors the innovations in the field of deep machine learning.