Nvidia’s latest AI can finally turn your crappy 3rd grade sketches into works of art
So you want to be a visual artist but your sketching, painting and other skills at creating visual images aren’t up to scratch? Well, very soon you will be able to fulfill your cravings with a little help from our artificial intelligence friends.
Chipmaker Nvidia has come up with a prototype of an app named GauGAN, an obvious wordplay on the French master Paul Gaugin and algorithms created by using generative adversarial networks – GANs that will be able to turn even the simplest of sketches into a standard painting/drawing interface and in a few seconds transform them into photorealistic pictures.
The process involves associating colours with objects that are to be portrayed in the prospective image – like green would be associated with grass. After the artist has done its part, the deep-learning part of the program that has been fed millions of images, will add the rest by itself. GauGAN will also feature a varying set of filters that can change day into night, adopt a certain painting style, and turn the initial sketched idea into a ‘real’ Gaugin or Ansel Adams.
As TechCrunch reports, since the program will be multimodal, you can tweak the same image with different color and style settings and let the artist become Andy Warhol creating a serial set. Even more so, if run on adequate gear, the results could be produced in real time, allowing the changes to be made instantly.
Sounds amazing for all visual art enthusiasts, but as MIT Technology Review points out, and Bryan Catanzaro, VP of Applied Deep Learning Research concurs, is the possible misuse of the system in creating strikingly realistic fake images, with Catanzaro giving assurances that the AI community is trying to tackle the problem on its part, but that this question needs to be fully tackled on a wider societal level.