The Future Is Here
We may earn a commission from links on this page

Watching This Neural Network Render Truly Photorealistic Faces Is Creepy and Mesmerizing

We may earn a commission from links on this page.

In 2015, Google released DeepDream, a bonkers, art-generating neural net users put to work rendering everything from disturbing dog collages to even more disturbing psychedelic porn. DeepDream may have just been the prelude to less aesthetically off-putting but much more significant applications of the slightly creepy technology—such as generating photorealistic, high-definition images of people who never existed.

Graphics card and electronics manufacturer Nvidia released a paper on Friday showing off a new method of generating unique faces via a generative adversarial network (GAN), a class of algorithm where researchers pair two competing neural networks against each other. In a GAN, one of the two neural networks is put to a generative function (like rendering images or trying to solve a problem) while the other is put in an adversarial role, challenging the first’s results. The intent is that the generative neural network will produce a superior result by bouncing its ideas off its adversarial counterpart.

Nvidia’s team wrote that with a new progressive training method, they were able to generate “images of unprecedented quality” using the CelebA-HQ database of photos of famous individuals, and that the results looked pretty good up to 1024 pixel resolution:

We describe a new training methodology for generative adversarial networks. The key idea is to grow both the generator and discriminator progressively, starting from low-resolution images, and add new layers that deal with higher resolution details as the training progresses. This greatly stabilizes the training and allows us to produce images of unprecedented quality, e.g., CelebA images at 1024² resolution.

It’s frankly pretty eerie, all these people that were never real floating past your eyes in uncanny detail—though it’s only static images, meaning we’re still pretty far off from the Matrix. Nvidia’s method also allowed for very good generation of objects and scenery, which you can see in the video below.

[Nvidia/Prosthetic Knowledge]