It’s an absolute nightmare: just a few clicks and your own face will be shown in porn without being asked. What once looked like poor Photoshop work is now almost indistinguishable from reality thanks to modern AI. It’s no longer just Hollywood stars like Taylor Swift or Scarlett Johansson who are affected. This is also happening more and more often to normal women who simply share pictures on social media. It is precisely at this point that IT forensic scientist Jens Kramosch takes over the digital search for clues.
Interview with IT forensic scientist Jens Kramosch
Jens Kramosch (42) is the founder of Lesk.red from Erfurt. His company tracks down deepfakes, analyzes their spread and supports those affected in taking action against manipulated content. In the interview, he reveals how he goes about it, what exactly it takes to create a deepfake and how we can protect ourselves from digital violence.
FOR YOU: What exactly is deepfake porn?
This is manipulated pornographic content in which a person’s face, body or even voice is inserted into a video or image without their consent. Those affected have nothing to do with these recordings, but to outsiders they appear as if they were involved
FOR YOU: How is manipulation done?
Deepfake porn is created today in two ways: A person’s face is often superimposed onto existing pornographic video material in a deceptively realistic manner – a kind of digital “face swap”. Such videos are increasingly being generated from scratch using AI. A few images and a suitable prompt are often enough to produce surprisingly realistic results.
FOR YOU: Can you explain the technology behind it in simple terms?
Essentially, this works via AI that is trained with images or videos of a person. The systems learn how a face looks and moves. This face can then either be transferred to existing material or incorporated into completely newly generated content. Often controlled via simple text input, so-called prompts.
FOR YOU: Why do the results look so realistic today?
The development was extremely fast. Just a few years ago, many AI videos appeared distorted or unrealistic. Today the models are so good that they can portray the finest details such as skin, light or facial expressions very convincingly. With the right prompts, it is now often difficult to distinguish between real and artificial material.
FOR YOU: What role do our public images and videos on Instagram & Co. play?
A central role. Publicly accessible images, for example from Instagram or Facebook, provide exactly the training material that the AI needs. Just a few photos can be enough. With more images, the result becomes even more precise. In many cases, this content comes directly from social media.
FOR YOU: Can anyone create content like this?
Basically yes. The barriers to entry have fallen significantly. Many tools today are designed to work with simple text input. The better someone formulates these prompts, the more realistic the result will be.
FOR YOU: Can such content completely disappear from the Internet?
This is very difficult. If a video is detected early, with a bit of luck it can still be removed. But once it has been shared and copied multiple times, it spreads quickly. And in the so-called backroom of the Internet, a lot of content remains permanently discoverable or reappears later.
FOR YOU: Does the problem only affect celebrities?
No, it’s no longer just celebrities. The barrier to entry has become so low that private individuals are increasingly being affected.
FOR YOU: What are the most common motivations of perpetrators?
The motives range from voyeuristic interests to targeted humiliation. In many cases it is about digital sexual violence or revenge, for example after relationships. We particularly often see cases following separations or conflicts, i.e. as targeted attacks against individual people. Technology is used to exert control and expose people.
FOR YOU: What can I do if I become a victim of a deepfake myself?
Acting quickly is crucial: secure evidence, i.e. save links and take screenshots, and report content directly to platforms. Additionally, legal action should be considered and, if possible, professional assistance should be sought to track the spread. The sooner you react, the better the spread can be contained.
FOR YOU: How can deepfakes currently be recognized and where do you reach your limits as an expert?
I often compare it to a crime scene: first you look at the big picture, then the details. Hair, hairline, eyelashes or skin structure are often noticeable. The skin often appears too smooth. It is also important to check the edges of the image. Because many AI models focus on the face and neglect the surroundings. At the same time, the fakes are getting better and better, and some are difficult to detect, even for experts.
FOR YOU: How does your company work to uncover such content?
We specifically detect deepfake content, analyze its distribution and document where it first appears and how it spreads. We create a secure chain of evidence for those affected, which can also be used legally, for example by “freezing” content and documenting it forensically.
FOR YOU: You also use AI tools. How reliably do these tools work?
Yes, we use AI to detect AI fakes. These systems can detect patterns and inconsistencies that are barely visible to the human eye – for example at the pixel level. In the end, we always work with a combination of technology and human analysis to get reliable results.
FOR YOU: How can we protect ourselves from deepfakes?
There is no complete protection. Closed social media profiles can help, but not if the content or perpetrators come from your own environment. Basically, if you share less material publicly, you reduce the risk. However, it cannot be completely avoided at the moment.
FOR YOU: What is the current legal situation in Germany – are those affected adequately protected?
In my opinion there are still gaps. Although there are legal principles, such as personal rights, in practice it is often difficult to quickly identify perpetrators and have content removed quickly. The topic of digital violence is currently being discussed more, but the technology is developing very quickly.
FOR YOU: Will we even be able to distinguish between real and fake in the future?
It will be significantly more difficult. In the future, it will probably be more important to actively label real content. For example, with some kind of digital proof of authenticity. At the same time, I am optimistic that recognition technologies will continue to improve because every AI leaves certain patterns behind.
FOR YOU: How can we get a better handle on the problem?
We have to close legal gray areas and make platforms more responsible. At the same time, more awareness of digital violence is needed. The current attention to the issue is an important step, but regulation must keep up with the speed of technology.
