This Person Does Not Exist

The image below shows eight faces that were generated through a machine learning technique called a Generative Adversarial Network (GAN). The images were generated by the website (if that gives you a black page, try As the URL of that website suggests, these faces are completely artificial, and do not correspond to existing persons!

Persons that don't exist.Persons that don't exist.

Basically, a GAN is a combination of two neural networks. The first one (the generator) generates synthetic images, and the second one (the evaluator) tries to decide whether these images are fake or not. Initially, the evaluator is trained with a dataset of real images. After that, the two neural networks are trained as adversaries (hence, generative adversarial network). The first one tries to create images that fool the second one, and the second one tries to become better at spotting fake images. This technique seems to be able to generate some very convincing fake images.

The site was created to demonstrate the results of a new state-of-the-art technique for GANs. The academic paper that describes the algorithm is on arXiv at A Style-Based Generator Architecture for Generative Adversarial Networks, and there is also a very accessible web page with a description of how the algorithm works at Style-based GANs – Generating and Tuning Realistic Artificial Faces. I’ve embedded a YouTube video from that article below.

Apart from demonstrating the main feature of the novel technique, which is that it allows separately changing different aspects of the images (gender, age, etc.), the video also clearly demonstrates that these images are not at all generated by simply combining a few faces as you would do with image editing software.

And, yes, I have noticed that some of the faces on are sometimes off on a subtle or not so subtle level; in particular teeth and ears are sometimes a bit messed up… However, that doesn’t change that these state-of-the-art images are quite impressive, if you ask me.


Add new comment

The content of this field is kept private and will not be shown publicly.
Spam avoidance measure, sorry for this.

Restricted HTML

  • Allowed HTML tags: <a href hreflang> <em> <strong> <cite> <blockquote cite> <code> <ul type> <ol start type> <li> <dl> <dt> <dd> <h2 id> <h3 id> <h4 id> <h5 id> <h6 id>
  • Lines and paragraphs break automatically.
  • Web page addresses and email addresses turn into links automatically.
Submitted on 5 March 2019