NVIDIA AI Creates Damn Good Images of People It Made Up
Developments in Artificial Intelligence (AI) and Machine Learning (ML) have been advancing at an unreal pace this year. Using neural networks to create fake images is nothing new, but a recent paper published by NVIDIA researches shows just how far it has come. The three people in the images below are all fake! The resolution and quality of images that are being produced by generative methods have become scarily impressive of late.
NVIDIA’s Generative Adversarial Network (GAN) is behind these images and all were created by using the latest in style transfer technology. NVIDIA is now blending key characteristics from certain areas of the face to come up with real looking images of people that are not real. GAN was setup to look at three basic styles coarse, middle, and fine styles and merge them at varying degrees to come up with images.
With the latest computer hardware and some impressive software we are finally able to create realistic AI faces! It took NVIDIA’s researchers just a week to train their model on eight NVIDIA Tesla GPUs to create the faces.
The paper also shows that they can create realistic images of cats, cars and likely any other object they want once the network has been trained. That brings some troubling questions to mind though. What will this mean for society? This means that almost any digital image could be faked with AI. AI isn’t perfect with facial symmetry just yet, but its only a matter of time until that is overcome. The next 10 years might radically change the modeling industry and even the porn business. It might be far cheaper to create images with AI than hiring real models for the work. It will likely also be used for propaganda for and against authorities. Just wait till we go from still images to animations!
New jobs will be created though as people will need to come up with ways to authenticate digital images, setup the hardware to make these networks and then write the software to properly train the network.
You can check out a video that shows how far NVIDIA has come below.