Various techniques to manipulate images was introduced in the 19 century and later applied to motion pictures. These techniques improved rapidly with digital video.
In the early 1990s, researchers at academic institutions developed Deepfake technology, which was later fine-tuned by developers in online communities. Recently, deepfakes have attracted a lot of attention for their uses in financial fraud, hoaxes, and fake news.
This has forced the government and industry to detect and limit their illegal use. However, the technology has a lot of potential if used in the correct way.
What Is Deepfake Technology?
Deepfakes refer to manipulated visual content generated by sophisticated artificial intelligence, which yields fabricated pictures and sounds that appear to be real. In most cases, an individual in an existing video or image is replaced with someone else’s likeness.
Deepfakes are created by using deep learning models — a subclass of machine learning methods based on artificial neural networks with representation learning. It involves training generative neural network architectures like generative adversarial networks or autoencoders.
Although it is difficult to create a good deepfake on a conventional computer, there are plenty of tools available on the internet to help people make decent deepfakes. The technology is still in its infancy, so don’t expect the perfect output.
We have carefully gathered a few good deepfake apps and tools that do not require a high-end desktop with powerful graphic cards (except one or two). You can use them either for research purposes or just for fun, but don’t violate anyone’s privacy.
Doublicat lets you take a selfie and put your face on to a meme or GIF in its library. It takes about 5 seconds for your face to be overlaid on that of Brad Pitt, Leonardo Dicaprio, or Taylor Swift.
You will be surprised after seeing how well your overlaid face adopts the same expressions as the original. You can forward the results to your family and friends, or post it to Instagram.
The output will be quite weird if there is a lot of face movement, but overall, it is an interesting experiment. As per the app developers, the image itself is deleted from the servers right after it is processed. However, it saves representations of facial features.
FaceApp is developed by a Russian company Wireless Lab. It uses neural networks to generate highly realistic transformations of faces in photos.
The app can transform your face to make it smile, look older, look younger, or just have fun with gender swap, along with many other exciting transformations. Tattoos, vignettes, lens blur, and background overlays are also a part of FaceApp.
In 2018, the app attracted a lot of attention from the transgender and LGBT communities because of its realistic gender-change transformations. It has also faced criticism on both social media and press over the privacy of user data.
4. Deepfakes web β
Price: $2 per hour
With this tool, you can create deepfake videos on the web. However, the learning curve here is little more than what you would find in other apps.
You need to signup and upload your videos. Everything else happens on the cloud that uses powerful GPUs. It takes nearly 4 hours to learn from video/images and swap faces. You can also use the trained model to swap faces., which takes about 30 minutes.
The output video quality depends on the ‘loss’ values: the lower the loss values (while learning from uploaded videos) the higher the quality. And of course, only you can access your videos and learning data.
Deepfake quality progress
DeepFaceLab is a leading software for creating deepfakes. It utilizes novel neural networks to replace faces in videos. It is hosted on GitHub and has spawned countless tutorials on the internet.
DeepFaceLab works great, but you need to have the technical knowledge to use it. Once you download and unzip the tool, you will see numerous folders and a series of batch files. There is a folder named ‘workspace’ that consists of all training models, source videos, and the output. The tool works with specific file names and locations so that the batch file can function.
Platform: Windows | MacOS | Linux
FaceSwap is similar to DeepFaceLab, but it provides more features, better documentation, and better online support. And yes, it is also available on Mac and Linux.
It’s an open-source tool packed with functionality to perform every step of the deepfake process, from importing initial videos to producing a final deepfake video. To run this tool, you need to have a powerful graphic card(s), as face-swapping on CPU is incredibly slow.
Powered by Python, Keras, and Tensorflow, Faceswap has an active community supporting and developing the software. There are plenty of tutorials to help you get started.
Zao’s deepfake technology allows you to modulate the voices of celebrities and stitch your face onto an actor’s body in a scene.
Just click one picture and try thousands of trendy hairstyles, clothing, and makeup. The app gives you tons of video clips, outfits, and literally unlimited possibilities to explore.
It takes only a few seconds to swap your face, but since the algorithm is mostly trained on Chinese faces, it may not look as natural as you would expect.
Nevertheless, all these tools demonstrate how quickly the underlying AI has evolved: what once required thousands of pictures to make a rather convincing deepfake video now requires just a single picture and yields better outputs.