Just over a year ago, we talked about NVIDIA and its AI processing technology that worked to create realistic human faces from some source photographs. The pictures that AI created was so photorealistic that you would be hard pressed to tell the difference between AI generated faces and real human faces in pictures. Technology has come so far over the past four years, that the difference now is even more amazing, and frankly a bit creepy.
The image above (b) shows the faces that a similar AI could generate in 2014, while the image under this section shows faces AI can create today. The AI-rendered faces are so good that if you were not told that they were manufactured by a computer, you would probably assume they were real people. These AI-generated faces promise to launch a new generation of photorealistic human representations in video games and movies, without the need for human actors or extras.
NVIDIA researchers changed a basic Generative Adversarial Network or "GAN" architecture to create the photos in the latest research (PDF). GAN's works on the basis of a network of networks that are working with a discriminator network to detect whether data generated (by the generator) represents an authentic version. The faces generated by AI are said to be easily customizable using a method called style transfer that blends the properties of one image with another. This is similar to how some apps on Facebook work, like those who do the pictures, look like impressionistic paintings. When the style transfer is used, AI can generate images that look like completely different people.
When AI gets so good at making fake pictures, some fear that technology can destroy the trust people place in pictorial documentation. Researchers working with this kind of technology point out that you can not change an image you want with the same fidelity. Making the mistakes takes time; The NVIDIA team says it took a week to train its model powered by eight Tesla GPUs (basically a small AI server / supercomputer). Human hair is said to be the most likely giveaway of an AI-generated image because it is hard to fall and often looks painted on. You can see it in some of the NVIDIA-generated images, but some of them look almost perfect. If the AI-generated images have reached as far as four years, people who criticize the technology must wonder what the mistakes can be for four years and where this will lead us in the future.