Contents:

Master neural networks without spending money: "Practical course"
Learn moreYou've probably seen videos of Tom Cruise masterfully playing the guitar and demonstrating card tricks, or recordings of Barack Obama calling his successor "completely stupid." These videos look surprisingly believable, despite being completely fictitious. This phenomenon is called deepfakes—digital manipulations that become so believable that even experienced viewers sometimes fail to recognize them as fake.
A few years ago, creating such videos required not only professional skills but also expensive equipment and weeks of hard work. However, today, thanks to advances in neural networks, almost anyone with a smartphone can insert their image into a famous Hollywood film or "revive" a photo of a deceased relative in just a few minutes.
In this article, we will discuss in detail what deepfakes are, how they work, and where they are used. We'll also explore tools that allow you to create your own deepfakes and offer a step-by-step guide to creating your first deepfake, from choosing source material to finishing the final edit.
Contents
- Deepfakes are an artificial intelligence-based technology that allows you to create fake video and audio recordings. This is achieved through deep learning algorithms that analyze and process images and sounds to replace the faces or voices of real people with others.
Deepfakes are based on neural networks that are trained on large amounts of data, including images and videos of the original characters. The system can then generate new frames in which the faces are replaced, and facial expressions and movements are adapted to the desired style. This can create the illusion that a person is saying or doing something that did not actually happen.
While the technology has its positive aspects, such as its use in film and animation, it also raises serious concerns. Deepfakes can be used to manipulate public opinion, spread disinformation, or create compromising materials. This raises important questions about the ethics and responsibility involved in using such tools.
- Deepfakes have applications in a variety of fields. They are primarily used in the entertainment industry, for example, to create special effects in films or to recreate the images of actors. Deepfake technology is also actively used in advertising, allowing for the creation of more engaging and memorable commercials.
Furthermore, in recent years, there has been a growing interest in deepfakes in the educational field, where they can serve as a tool for creating interactive educational materials. In journalism and media, these technologies can be used to create parodies or satirical videos, although this also raises ethical questions.
Despite the positive aspects, there are also negative applications of deepfakes, such as the creation of false information or the manipulation of public opinion, which raises serious concerns among experts and society at large. Thus, using deepfakes has both benefits and risks that must be considered. The legality of creating deepfakes depends on a variety of factors, including jurisdiction and specific circumstances. In some countries, the use of deepfake technology may fall under copyright, privacy, and defamation laws. For example, if a deepfake is used to damage a person's reputation or to deceive, this could lead to legal consequences. At the same time, there are cases where creating deepfakes may be legal, particularly if used for entertainment purposes or with the consent of all parties involved. It is important to keep in mind that the legislation in this area continues to evolve, and many aspects remain uncertain, making the question of the legality of deepfakes quite complex and multifaceted.
- How to spot a deepfake
- The best tools for generating deepfakes
- Making a deepfake yourself: a step-by-step guide.
Deepfakes: the essence of the technology and how it works
A deepfake, derived from the English term deepfake, is a synthetic media file, be it video, audio, or image, which is created using neural networks. This technology makes it possible to replace the face or voice of one person with the face or voice of another. In other words, a deepfake is a method of portraying a person in circumstances in which they were never actually present.
Deepfakes became popular thanks to viral content. A striking example is the TikTok series of videos featuring "Tom Cruise," in which the virtual character performs various stunts, leading many users to believe he's real as the Hollywood star. Also worth mentioning is the controversial video featuring "Barack Obama," in which the former US president seemingly expresses harsh opinions about his successor. These videos are striking in their realism, but in fact they are the result of neural networks at work.
On the technical side, most deepfakes are based on two key machine learning technologies:
- Generative adversarial networks (GANs) operate based on the concept of competition between two neural networks. One, called the generator, is responsible for creating images, while the second, the discriminator, seeks to figure out whether a given image is genuine or artificially generated. Through this competitive process, the generator gradually improves its skills, producing images that become more believable and harder to distinguish from the real thing.
- Autoencoders are a special class of neural networks that initially "compress" an image, converting it into a set of key features, and then reconstruct the image from these features. In the process of creating deepfakes, two such autoencoders are used: the first one is trained on images of a face that acts as a source, and the second one is trained on images of a target face. After this, the encoded characteristics of one face are combined with the decoder of the other, which effectively "superimposes" one face on another.

Read also:
In this article, we will discuss how artificial intelligence creates images of cats using generative adversarial neural networks, known as GANs. These neural networks are a powerful tool that allows algorithms to learn from large amounts of data and generate new images that can look quite realistic.
Generative adversarial neural networks operate on the principle of interaction between two components: a generator and a discriminator. The generator creates new images, attempting to imitate real ones, while the discriminator evaluates how similar the created image is to the real one. This process resembles a competition, where both elements constantly improve their skills. The generator strives to fool the discriminator, and the discriminator, in turn, tries to recognize fakes.
Thus, as the GAN trains, the generator becomes increasingly adept at creating cat images, and the discriminator at distinguishing them from real photos. This leads to the creation of increasingly high-quality and realistic images. These technologies open up new horizons for creativity and design, enabling artificial intelligence not only to reproduce but also to create unique visual images.
To produce a high-quality deepfake, neural networks need to process a large volume of images, both from the "donor"—the person whose face will be used—and the "recipient"—the person whose face will be replaced. The more diverse these images are, taking into account different camera angles, lighting, and facial expressions, the more natural the final result will be.
Modern technologies make it possible not only to swap faces in images but also to reproduce facial expressions, lip movements that synchronize with speech, and imitate the human voice. With the development of artificial intelligence, it is becoming increasingly difficult to distinguish deepfakes from real video or audio recordings.
Applications of deepfake technology
Deepfake technology is used in various areas of our lives. Let's take a closer look at the key areas where it is used:
- The entertainment industry actively uses deepfake technology, and film studios and agencies integrate it into their projects. For example, in the film "Rogue One: A Star Wars Story." Stories" digital versions of the young Princess Leia and Grand Moff Tarkin were created using modern computer technology.
- Advertising agencies are using deepfake technology to create viral videos featuring famous people.
- In the field of education, history museums are developing interactive exhibitions, allowing visitors to "interact" with prominent figures of the past.
- Medicine and scientific research. Deepfake technology contributes to the training of medical specialists, allowing the creation of realistic simulations of various patient conditions and the illustration of rare diseases. The use of such technologies makes scientific visualization of complex processes more accessible and understandable. In addition, deepfakes are used to develop virtual assistants designed for patients with cognitive impairments.
Legal aspects of deepfake creation
Legislative norms concerning deepfakes have not kept pace with the rapid progress of technology. In Russia, there is no special law that would directly regulate the process of creating and distributing synthetic images and voices. However, existing legislation already provides mechanisms for protecting personal rights.
For example, Article 152.1 of the Civil Code of the Russian Federation guarantees citizens the protection of their rights to their own image. The use and publication of such images is permitted only with the consent of the individual, except in situations related to state or public interests.
A striking example of the violation of rights to participate in elections was the incident involving Russian voice actress Alena Andronova. In 2023, she learned that her voice was being used in advertising materials containing inappropriate content.
Experts confirmed that the voice was created using a neural network based on recordings of Andronova's speeches. The artist initiated legal proceedings to protect her rights, and this case became one of the first in Russia concerning the illegal use of a synthesized voice.
There are many situations in which creating deepfakes would clearly violate the law:
- Fraudulent activities. Faking a video call from a loved one or acquaintance with a request to transfer funds. This deception scheme is already being used in real-life cybercriminal practices.
- Disinformation is false information that can take the form of fake news or statements attributed to politicians, entrepreneurs, or public leaders. The main goal of such actions is to manipulate people's opinions and influence developing events.
- Non-consensual pornographic materials are one of the most common and traumatic uses of deepfake technology. In such cases, a person's face is placed in explicit videos without their permission. This can lead to serious consequences, including damage to reputation and psychological trauma.
- Slander and defamation are actions aimed at spreading false information that can undermine a person's reputation and worsen their public image. Creating fictitious facts that can negatively reflect on the honor and dignity of an individual is a serious offense.
- Personal data leakage. Using a fake voice or image to bypass a biometric authentication system in order to access protected information.
How to spot a deepfake
Although algorithms continue to evolve, many deepfakes still have obvious flaws. The most noticeable signs of a fake are unnatural blinking and eye movements. The average person blinks 15 to 20 times per minute, while deepfakes may blink either too rarely or too often.
One common sign of a deepfake is problems with facial boundaries. If you look closely, you can detect blurring, flickering, or distortion at the edges, especially when the person turns their head or speaks. These visual anomalies arise from limitations of the algorithms used to superimpose one face onto another.

To more effectively detect deepfakes, experts have created a number of specialized programs and services. For example, Microsoft Video Authenticator analyzes video files and calculates the likelihood of their modification, identifying subtle signs of forgery. Additionally, there's an online service called Deepware Scanner, which also evaluates video materials and provides a probabilistic assessment of their authenticity.
In addition to using technology, it is also necessary to improve your skills in critically perceiving media materials. Be sure to evaluate the credibility of the source of information: if the video is posted on a little-known site or in an anonymous channel, you should be especially careful.
The Best Deepfake Generators
Today, the market is filled with a variety of deepfake development tools. Here you can find both professional applications that require significant technical skills and mobile programs available to every smartphone owner. Let's consider the most common and effective solutions for generating synthetic media content with varying degrees of complexity.
Operating systems: Windows and Linux
Cost: free
DeepFaceLab is the benchmark in deepfake development. Approximately 95% of all deepfakes found online were created using this program.
DeepFaceLab's main advantages are the outstanding quality of the resulting videos and its extensive customization options. The software offers a variety of models for training, as well as various algorithms that allow you to customize the process of creating deepfakes depending on specific requirements and characteristics of the source data.

While DeepFaceLab has many advantages, it There are some drawbacks, too. For inexperienced users, the program may be difficult to learn, and it requires a high-performance computer—the more powerful, the better. For example, the minimum system requirements state a graphics card with 8 GB of RAM. However, with this amount of memory, creating short videos can take several days.
It should be noted that this project has been moved to the public archive since November 2024. This means that its developers are no longer providing updates. The program may have dependency issues that will require examining the DeepFaceLab source code to resolve.
Operating systems: Android, iOS, and web applications.
Price starts at $13 per month, with a trial plan available.
Reface is a mobile app that allows you to quickly and easily create deepfakes directly on your smartphone.
Reface works quite simply: you take a selfie or upload a photo from your photo gallery, then choose from a selection of videos or GIFs to embed your image in. In just a few seconds, the app creates a ready-made deepfake that can be instantly shared on social networks.

Reface demonstrates high speed of operation and does not require powerful system resources for its functioning, but the quality of the resulting videos leaves much to be desired. Additionally, the app has limited customization options.
Operating systems: Windows, Linux, and macOS.
Cost: Free
FaceSwap is a more accessible option for beginners than DeepFaceLab. This program has a simple interface, and the deepfake creation process is automated. As a result, users receive video materials in excellent quality.

Like DeepFaceLab, FaceSwap requires a high-performance computer equipped with a high-quality video card, but in general its resource requirements are less stringent. This program is available for download completely free of charge on the GitHub platform, and its development is actively monitored by a community of both developers and users.
FaceSwap is ideal for projects of moderate complexity, when the required quality exceeds the capabilities of mobile apps, but professional-level realism is not required.
Platforms: Web
Price starts at $15 per month.
Runway Act-Two is a neural network designed to "animate" various characters. To begin, you need to upload a video of a person performing movements and speaking words, and then provide an image of the desired character. Machine learning system will be able to transfer human movements to a static image of a character, thereby animating it.
The neural network operates in browser mode, which allows the user to be independent of high-performance devices. However, there is one limitation: videos created cannot exceed 30 seconds in length.
Platform: Web
Price starts at $8 per month, with a trial plan available.
Deep Nostalgia is a platform that uses deepfake technology to bring archival images to life. Unlike many similar services, which are primarily aimed at entertainment or generating inaccurate content, Deep Nostalgia offers users the opportunity to develop an emotional connection with their story.
The system processes the uploaded image, recognizes the human face, and uses pre-designed animation templates to generate a short video. In this video, the person blinks, turns their head, and makes slight facial expressions.

Making a Deepfake Yourself: A Detailed Step-by-Step Guide
To start using specialized software for creating deepfakes, you need powerful hardware and knowledge of machine learning algorithms. To make this process more accessible, we'll turn to the Act-Two neural network, developed by Runway. This neural network runs in the browser and allows for the rapid creation of high-quality videos.
Access to Act-Two is paid, but the cost is lower than purchasing a high-performance graphics card for local neural network use. The Runway platform offers the following pricing plans:

The basic plan for creating deepfakes is called Standard and costs $15 per month. Higher plans allow you to create more videos and provide greater storage capacity for assets on Runway's servers.
The organization does not accept Russian bank cards, so you must use international friends or bank accounts opened abroad for payments.
After completing the payment, you'll be redirected to the main interface of the Runway web app. Since we're working with deepfakes, we'll need to switch to Video mode. This switch is located at the top of the screen.

In the lower section of the window intended for creating a video, select the Act-Two option. This mode is just right for character animation, which is our priority.

The Act-Two editor includes several components:
- In the «Performance» section, you need to upload the source video, which contains the desired movements and speech. It's important that the person in the recording is shown full-face and clearly visible. Videos with dark or overly bright areas are unsuitable. To record a video directly in the browser, you can use the "Record" button.
- Character is the window for the character. You need to upload the image we'll be animating. The character in the picture must be presented full face.
- Voices is the section that allows you to select a voice.

We present you with step-by-step instructions for creating character animation in the game Runaway. To respect copyright, we will focus on animating a fictional character, whose image was developed using ChatGPT. Here is his description:

For animation:
- Step one. Upload the video to the Performance section. This video should capture your actions. Speak and gesture as you would like your future deepfake to act. Please note that the maximum video length is 30 seconds.
- Step 2. Open the "Voice" menu and select a suitable voice for your character. Runway offers a vast collection of voices, each with associated gender, age, and emotional state. We settled on Brody, the voice of a young and active man.

- Step 3. Click the Generate Voice button and wait for the process to complete. This usually takes about one minute. Once the generation is complete, you'll see a video of your modified voice on the right side of the screen, which you can save.
- Step 4. In the Performance section, upload the video with your changes. Then, select Character and add an image of your character. After that, click the Generate Character button.

- Step 5. Once the creation process is complete, you will see your deepfake on the right side of the screen, which you can save to your computer's hard drive.

This is what we ended up with:
What's the Bottom Line
- A deepfake is a video or audio file in which a real person is depicted in situations or saying words that they did not actually say or do.
- Deepfakes are created using specialized neural networks that superimpose the desired movements onto images of real people.
- This technology is used in various fields, such as the film industry, the advertising business, and educational institutions.
- Attackers also use deepfake technologies. For example, they can create videos featuring victims' loved ones to coerce them into transferring funds or disclosing personal information.
- In Russia, there are no regulations that directly address the development and use of deepfakes. However, according to the Russian Civil Code, posting photographs and videos of a person without their permission is unacceptable.
You'll find a lot of fascinating information about programming in our Telegram channel. Join us!
Read also:
- A neural network is a model inspired by the structure and functioning of the human brain, designed to process and analyze data. It is based on artificial neurons that are combined into layers. Each neuron receives input data, processes it, and passes the results to the next layer.
A neural network begins with information being fed to the input, after which the data passes through several layers, where each subsequent iteration allows for the identification of more complex patterns and connections. During training, the neural network analyzes the results of its predictions, comparing them to known answers and adjusting its parameters to improve accuracy.
Using algorithms such as backpropagation, the neural network adjusts its weights to minimize the difference between predicted and actual values. Ultimately, through a large number of examples and iterations, the neural network becomes capable of performing a variety of tasks, including image classification, natural language processing, and more.
- Top 12 Free Neural Network Tools for Image Creation and Editing.
- PyTorch Basics for Python Developers

