Deepfake: What It Is, How to Detect It, and How to Protect Yourself

Deepfake (from deep learning and fake) is a photo, video, or voice recording in which a neural network has replaced one person’s face, facial expressions, or speech with those of another person. The result looks as if the person really said or did something that never actually happened.

How Deepfakes Work

Neural networks are shown hundreds of photos and video clips of two people. They learn to break a face down into features—shape, lighting, head rotation, and lip movements—and then reconstruct it with someone else’s appearance. For voice cloning, just a few minutes of recording are enough: the model memorizes the speaker’s timbre and intonation and then uses them to voice any text.

Previously, this required a powerful graphics card and days of training. Today, online services can swap a face in a short video within minutes, so deepfakes have become widespread and their quality is improving rapidly.

Where Deepfakes Are Used Legally

The technology itself is neutral. With the consent of the people whose faces and voices are used, it can be applied:

• in film and advertising — to de-age actors, re-record lines without new filming, and dub films with lip synchronization;
• in education — to “bring historical figures to life” in museums and educational videos;
• in business — to create videos with a presenter’s digital avatar in multiple languages;
• in medicine — to restore the voices of people who have lost them after an illness.

Why Deepfakes Are Dangerous

Problems begin when someone else’s face is used without permission:

1. Fraud. Criminals call using the “voice” of a relative or manager and ask the victim to transfer money urgently.
2. Disinformation. Fake statements by politicians and celebrities spread across social media faster than retractions.
3. Non-consensual intimate fakes. A real person’s face is inserted into explicit content—this constitutes harassment and violates the right to one’s image. In some countries, such materials carry criminal liability.
4. Erosion of trust. When anything can be faked, genuine evidence also comes under suspicion.

That is why our catalog does not include services that replace people’s faces without their consent. In the European Union, the AI Act requires generated content to be labeled, while major platforms remove deepfakes that mislead viewers.

How to Recognize a Deepfake

There are no perfect indicators, but certain details often reveal a fake:

• unnatural blinking or a complete lack of it, along with a “glassy” stare;
• a blurred facial boundary around the hair, ears, and neck, especially when the head turns;
• teeth and the tongue merging into a single blotch, or lips not matching the audio;
• lighting and shadows on the face not matching the lighting in the scene;
• a voice that sounds flat, without breathing or pauses, or has a metallic quality.

If a video triggers strong emotions and urges immediate action, look for the original source: the person’s official account or a publication in reputable media.

How to Protect Yourself

• Agree on a code word with your family for “urgent” calls involving requests for money.
• Call back yourself using a known number rather than the number from which the call came.
• Do not make large amounts of video and voice recordings publicly available unless necessary.
• If you find a deepfake featuring you, save the links and screenshots and report it to the platform where it was published; if there are threats or extortion, contact the police.

Legal tools for working with video and voice—editing, animation, and voice-over using your own voice—can be found in the “Video and Animation” and “Audio and Voice” catalog sections.