New Deepfake App Allows Users to Superimpose Their Face on GIFs
Sun, April 18, 2021

New Deepfake App Allows Users to Superimpose Their Face on GIFs

The Doublicat app allows users to superimpose their faces into popular GIFs using deep learning technology / Credits: Mihai Surdu via Shutterstock

 

Last year, a fake video of Nancy Pelosi, Speaker of the US House Representatives, circulated all over social media. The viral video showed Pelosi slurring her words as if she was intoxicated. This is a perfect example of a deepfake, manipulated media created through the power of artificial intelligence. This form of disinformation could be used in several ways. But, deepfakes can mainly be used as a tool for presidential elections. It is predicted that deepfake videos will be used “to portray candidates saying and doing things they never said or did.”

Fortunately, social media sites are doing their part to stop deepfakes from spreading. For instance, Facebook announced that any deepfake video that is designed to be misleading will be banned. While many are thinking of solutions to prevent more deepfakes from being created, some companies are using these to introduce services or features.

Recently, RefaceAI introduced an app called Doublicat that allows users to superimpose their faces onto popular GIFs using deep learning technology. According to Artificial intelligence News, an online site that features AI bots, deep learning, machine learning, neural networks, and much more, Doublicat uses a Generative Adversarial Network (GAN) to swap out your face in a GIF. It works by simply taking a selfie and selecting a GIF to replace the original face with your face. Users can take selfies with different looks and accessories to experiment.

Roman Mogylnyi, the CEO of RefaceAI, stated that Doublicat doesn’t rely on the 3D model of the user’s face like popular face swap apps such as Zao. “In most of the deepfakes, they use the approach that we put to rest more than a year ago. I’m talking here about 3D masks, face reconstruction, and texture blending. This solution can work only with previously collected and preprocessed content. And can be unstable in facial expressions and turns,” he said.