Deepfake Technology: How to Create a Deepfake Online
The rapid development of technology allows users to perform tasks that seemed impossible 3-5 years ago. For example, they can change an actor’s face in a video and apply one method multiple times.
In this article, we will discuss the deepfake technology. We will explore how to create a deepfake for free using various services and highlight the mistakes to avoid. We will provide examples and give advice to beginners.
What is a deepfake in simple terms?
Deepfake is a technique of image manipulation based on artificial intelligence and machine learning. It uses AI tools to change the faces of characters in original images or videos.
Deepfakes emerged in 2017 when several videos with adult content were posted on a forum, replacing the faces of girls. Shortly after, numerous libraries and online services for creating deepfakes appeared in the public domain.
While video editing tools existed before, it was after 2017 that free tools became available, allowing any user to perform similar tasks.
Deepfakes are actively used in marketing in general and in affiliate marketing specifically. In the former case, they are created to generate viral reach.
Using deepfakes as the basis for creativity without the permission of the rights holder is not possible. If, for example, AT&T wants to create a video featuring a fake Elon Musk, the real billionaire must provide permission.
In affiliate marketing, deepfakes are mainly used for advertising in somewhat legal and illegal niches. They are particularly active during major events such as Apple’s annual September presentation.
You can find numerous examples online where Tim Cook, Mark Zuckerberg, and other opinion leaders urge people to invest in new tokens or investment projects, promising high dividends. These videos are generated using deepfake algorithms. However, they are less commonly seen in targeted creatives.
What is the basis of deepfake technology?
Deepfakes are created using neural networks and machine learning. Imagine being shown 1,000 photos of Elon Musk in a few minutes and being asked to recreate his portrait. It would be difficult for a person without drawing skills, but AI can handle it easily.
The fundamental algorithm behind deepfakes is based on Generative Adversarial Networks (GAN). They learn the facial expressions of a character based on thousands of photos and then create a new video with a different face.
The autoencoder method is also actively used. One algorithm learns to encode the original face, while the other learns the face to be superimposed on the video sequence. As a result, users obtain the desired material.
The use of GAN and encoder techniques generally yield the desired results, but this technology is still not perfect. It overlays the face onto the original character, but the forehead, hair, and other body parts remain unchanged. Therefore, visual errors can often be seen during motion.
Although AI learns from a large amount of data, sometimes it is not enough to create a realistic video. Similar problems arise with neural networks used for enhancing photo quality. Instead of increasing the resolution, the output image may become pixelated.
Simple tools for creating deepfakes provide surface-level processing, but there are more powerful tools with advanced features. However, using these tools effectively requires time for learning and training.
While there are many free tools available online for creating deepfakes, it should not be expected that they will produce perfect results. The outcome depends on the parameters of the source video. If the video is too complex for AI, it may not be possible to obtain a realistic image.
Examples of photo and video deepfakes
You can find them on YouTube, for example. They are used for entertainment purposes or to achieve specific goals, such as increasing click-through rates on Facebook.
Beginners can easily find free tutorials describing the process of creating deepfakes. The final result depends on the specific project’s characteristics and the quality of the source materials. Some videos may not turn out as beautiful as desired, requiring multiple iterations.
There is an Instagram account impersonating Elon Musk. It is managed by a neural network that overlays Musk’s face on photos of random women. The account has 57,000 followers, and each post receives a minimum of 10-20 comments.
The example of a deepfake featuring Mr. Bean as President Trump has garnered over 2 million views in 3 years. However, the quality of the video is not high, and the original facial features become visible when the character’s position changes.
Another edited video featuring Mr. Bean, this time with the actor’s face superimposed on a female body, resulted in a simultaneously cringe-worthy and captivating advertisement. The video has gathered 1.3 million views.
The next deepfake turned out to be highly realistic. Jim Carrey was considered for the lead role in the movie “Joker” but it ultimately went to another actor. Using neural networks, the developer showcased how the film could have looked with Carrey in the role.
The video featuring Jim Carrey is very realistic. This is due to the meticulous editing of frames and the fact that the character remained stationary. It is much more challenging to replace a face when a person is in motion.
These examples of deepfakes demonstrate that it is easier to change faces in photos. Neural networks only need to analyze a few images to successfully accomplish the task.
How to create a deepfake online: 5 services
Some sources claim that anyone, including beginners, can create a deepfake in 30 minutes. They suggest visiting the first available website from search results, uploading photos, and waiting for the content to be processed.
In reality, online services are not suitable for creating high-quality deepfake videos. Free tools may handle simple tasks like animating photos, but when it comes to creating high-quality advertising creatives, powerful software is required.
Web-based platforms that have created numerous deepfakes typically use desktop scripts with graphical interfaces capable of performing a large number of computational operations. These tools provide the best results.
Almost every deepfake creation service can handle basic photo or video editing, but the available features are unlikely to be sufficient for obtaining several worthy options for use in advertising campaigns.
This mobile application excels at replacing faces in photos, GIFs, and videos. To edit the content, you need to upload a person’s face and choose from the provided templates.
The output can be a photo or video of average quality. The application has a simple interface, but the algorithms do not always handle face replacement perfectly. The free version includes a watermark on videos.
Is a highly popular application that gained significant traction on social media. It can alter people’s appearance in photos by applying effects such as aging, rejuvenation, face swapping, and hairstyle changes.
Once you upload a photo, you can utilize all the available effects and built-in presets. FaceApp excels at facial editing, and smile transformations and other changes in facial expressions look flawless.
Is a script that has been used to create over 90% of all deepfakes, according to research. It can be used on Windows and Linux computers. To process a large number of source files, a system with at least 2-4 GB of RAM is recommended.
Deepfacelab enables the creation of high-quality deepfakes, but it requires time to learn the program’s intricacies. Most operations are performed through the command line.
Is a script that specializes in static images and can change faces in photos. Even from demo examples, it is evident that the algorithms perform their tasks exceptionally well. However, FaceShifter does not work with videos; it can only be used to create images. There are few worthy competitors in this niche, making FaceShifter worth downloading.
Below is a list of services and software, their platforms, prices, and key features:
|Reface||Android/iOS||$33 lifetime||built-in templates, one-click deepfake creation, fast face recognition|
|FaceApp||Android/iOS||$19 per year||built-in templates, background changes, hairstyle and facial editing|
|Deepfacelab||Desktop||free||head shape alteration, lip manipulation, high processing speed|
|FaceShifter||Desktop||free||face replacement based on 1-2 photos, fast content processing|
While some tools on the list may not be considered full-fledged deepfake generators, each of them possesses useful functions. In some cases, it is simpler to use an Android app than delve into command-line scripting.
Experienced individuals can utilize Python programs or cloud-based systems, while newcomers may find Deepfakes Web service a viable option, providing relatively good results for a modest fee of $2.
Why Affiliates Use Deepfake Photos and Videos
They can be used to create Facebook or other social media creatives for advertising campaigns and testing on a small amount of traffic.
Deepfakes featuring celebrity faces rarely pass moderation. Sometimes, text, audio tracks, and other parameters need to be altered to obtain approval from advertising network algorithms.
Regarding creative freshness, deepfakes no longer astonish users as much as they did 1-2 years ago. They can be tested on specific GEOs, such as Indonesia or Egypt, but local influencers should be used.
When combined with a text-to-speech neural network, deepfakes can save costs on voice-over work. Recording voiceovers in a studio for several videos can cost several hundred dollars, whereas an AI can do the job for free.
Deepfakes are intriguing tools that can find applications in advertising, marketing, and affiliate marketing. By using services and software correctly, one can maximize their benefits.
For affiliates who want impressive deepfakes for creatives, it is advisable to entrust the task to experienced developers. They can create higher-quality videos that have a chance of bypassing the algorithms of Facebook and other social media platforms driven by big data.
How does deepfake technology work?
The technology is based on neural networks that replace the original face with an image of another person. AI analyzes facial features, expressions, and edits the video footage.
Is programming skills necessary to create a deepfake?
For creating simple ones, programming skills such as Python are not necessary. However, powerful scripts like Deepfacelab work through the command-line interface, requiring some programming knowledge.
Which services can be used to create deepfakes?
There are dozens of applications, programs, and scripts without a graphical user interface available online. Among the popular options are Reface, FaceApp, Deepfacelab, and FaceShifter.
Are any of these tools free to use?
Most smartphone applications come with paid versions that offer additional features and remove watermarks. However, desktop scripts are generally free to use.
Can deepfakes be created using Python?
Yes, they can be created using Python, but it requires specific libraries and knowledge of the programming language. Beginners may find it easier to use simpler tools.