Source - Avatarify input & output
Are you using Zoom or Skype too much during lockdown days to connect to your friends? Getting bored often? Then you can try deepfake program Avatarify. It uses artificial intelligence to make a deepfake of your selected person. Deepfake is basically a face swap.
Image Source - First Order Motion Model for Image Animation
Avatarify is an open-source project. It is made by Ali Aliev & Karim Iskakov. He made this program using ‘First Order Motion Model for Image Animation’, developed by the University of Trento researchers. The image animation model generates a video sequence so that an object in the image source gets animated according to the motion of a driving video. The source code of Avatarify is available in Github. You just need one photograph of the person to build an avatar of that person. A high-resolution photo gives a better result. With the help of this program, you can literally be anyone under the sun. You can be Elon Musk, Einstein or Trump. The program generates streams to the virtual camera of the program. Your video chatting software like Zoom or Skype is not able to distinguish between real or virtual camera, so your avatar gets created virtually from the selected photograph and it gets aired during the video call. The program uses artificial intelligence and neural network model. A neural network gets trained continuously. More you train the model, the model becomes perfect. So if you are not satisfied with the output you can expect that the outcome will be improved down the line.
The developer Aliev creates a real-time deepfake of Elon Musk and enters into a Zoom call – From Aliev’s YouTube channel
At present, the program runs on GPU only and I'm not sure whether CPUs are supported or not. As GPU can render high-resolution pictures and videos concurrently, GPU is optimal for neural learning method. Supported operating systems are Linux, Mac OSX. Again I’m not sure whether it runs or Windows or not. It is not a plug and play program. You need to have some basic programming knowledge or command line knowledge to run the program.
Okay guys! Avatarify is made for fun. Maximum deepfake programs require recording of the video to make a convincing deepfake. Such programs take time to create a deepfake. This program creates deepfake on real-time basis. Yes, you need to train the model on the image for a good time before creating the deepfake obviously. You can enter into a Zoom call with any avatar now. Your friends won’t recognize you if they can’t find your voice suspicious. Wait a moment! There are tools to bypass that glitch also. Audio deepfake programs are also reality. You can do a Zoom call now with both video and audio deepfakes without giving any hint about your real identity. Have fun and don’t misuse it. If you wanna try the program, you can check the Github link given below.
The Github link of the project: https://github.com/alievk/avatarify