How Deepfake Works To Make Photos Or Videos Look Realist

Deepfake illustration (doc. Unsplash)

YOGYAKARTA What is deepfake is an artificial intelligence (AI) that is capable of producing images, videos, and fake audio. The term deepfake comes from deep learning and fake. The way deepfake works simply combines various digital data and is then adjusted to use AI. In making deepfake videos, AI will manipulate a person's facial expression by parents can also exchange faces as a whole.

Reporting from Tech Target, basically deepfake is not a video or edited image through edit tools such as photoshops. The making uses a special algorithm that combines old and new recordings. A person's facial feature will be analyzed using machine learning (ML), the analysis results will be used for manipulation activities in other videos.

There are two algorithms used, namely generators and discriminators. Both are used to create and perfect fake content.

The generator algorithm will initially build a training data set based on the desired output. While the discriminator algorithm will be used to analyze how realistic or fake the initial version of the content is.

The analysis process in this algorithm continues until AI is able to learn so that the resulting content is more realistic. Discriminators also continue to be skilled at finding weaknesses or shortcomings in content. The repair process continues until the content really appears real.

The combination of algorithms between generators and discriminators will create a generating adversarial network (GAN). GAN uses deep learning to recognize the pattern on the original image. After that the pattern will be used to produce new fake images. The GAN system will also collect details and various photo perspectives from various angles. The data will be applied to the new original content so that the results are even more subtle.

In the deepfake video, GAN will also do the same thing, as well as analyze behavior, movement, and speech patterns. The data will then be sold through the discriminator several times so that the video will be more perfect and realistic.

Please note that deepfake videos can be made in two ways, using the original video belonging to the target or exchanging faces in other people's videos. In the first way, the video will look like the original only audio has been changed as desired. While in the second way there will be manipulation of the audio and video.

Here are some specific things that will be done on deepfake videos.

If the content is taken from the original video, the neural network-based deepfake autoencoder will analyze the content. This is done to understand the existing attributes ranging from facial expressions to body language. After that these characteristics will be applied to their original videos.

For audio deepfakes, GAN will clone someone's voice then build a vocal model and then use the model to produce the sound the creator wants.

This technique is also used in deepfakes by mapping voice recordings to videos. As a result, someone who appears in the video deepfake will appear as if to say manipulated words.

That's information related to how deepfake works. Visit VOI.id to get other interesting information.