Random Face Generator

Create a face using our AI face generator











How to save the picture

  • Right-click the image.
  • Choose the command Save Picture As.
  • Use the Save Picture dialog box to find a location to save the picture.
  • You can rename the picture as it's saved to your computer's storage system.
  • Click the Save button.





Over more 1 Million Fake Faces

Face recognition systems have been trained to use a large number of images to create facial impressions of people by mapping the geometry of certain facial features. The most precise faces are created by the Generative Adversarial Network (GAN) of Nvidia, which uses deep learning techniques to create realistic portraits from a database of photos. All this seems innocent enough until you realize that the face that smiles at you is not real, but is generated by a neural network algorithm.

It uses an algorithm to display a single image of a person's face, and for the most part people's faces that don't exist look real. I have developed an application that puzzles you about how to distinguish a real face from a fake one. Every time you refresh the page, it will show you a fake GAN-generated face.

I used a 1 million fake face record of GAN-generated images in the application to ask you if you can distinguish a real face from a fake one, as well as the Kaggle and Utkface records from real images. A network creates fake faces and decides whether they are realistic by comparing them with photos of actual people. You then take a quiz to try to identify an image as the real face of one that GAN has created.

During training, the generator network takes random noise as input and produces pictures that are realistic images that can be distinguished from the training data set. The discrimination network is trained to determine what the images of a real person look like and evaluates its images based on how realistic the generator images are. The conditional generator is represented by the conditional GAN (AC-GAN) and stack GAN models which learn image characteristics and labels during training, enabling image generation under conditions of user-defined features.