Nvidia has found a novel (or creepy) way of generating photo-quality human faces. The graphics company used artificial intelligence to construct them out of celebrity images, and the results can be quite convincing.

Researchers at the company posted a video, which you can watch below, showing the vast collection of the synthesized human faces. Many look no different from photos of real people.

To create the images, Nvidia researchers used an AI algorithm called generative adversarial networks or GANs, according to a paper published last Friday.

A GAN essentially uses two neural networks, which are computing systems modeled on how biological brains actually work. Then it pits them against each other to complete a certain goal.

For the Nvidia project, one neural network was devoted to generating the life-like images. The other network acted as a critic; it flagged which photos were accurate or not. As more images were judged between good and bad, the GAN fine-tuned its approach.

It's not the first time GANs have been used to produce synthetic realistic photos. However, the Nvidia researchers found a way to improve the AI algorithm to create more detailed images.

Nvidia's approach used the GAN to first produce a low resolution image at 4-by-4 pixels. From there, it proceeded to build off that image, by adding more details, but at progressively higher image resolutions. That allowed the GAN to create what at first looks like a blurry square into a photo realistic image at 1024-by-1024 pixels.

Nvidia's algorithm pulled from a library of 30,000 images. The GAN worked for 20 days and relied on a single Nvidia Tesla P100 GPU, a chip used in supercomputing data centers.

Although the Nvidia researcher wrote that their new approach can produce "images of unprecedented quality," it still isn't perfect. Not all the images appeared accurate. Some had odd-looking necks, or mishapen chins and foreheads.

The researchers wrote that the algorithm will have trouble deciding between when an object should be rendered curved or straight. Images created at even higher pixel resolution will also look more fake. But it might not be long before computers programs know how to accurately produce convincing human images.

"There is also room for improvement in the micro-structure of the images," the researchers added. "That said, we feel that convincing realism may now be within reach."