Nvidia has shown off a new AI software that is capable of taking computer drawn doodles and turning them into photo-realistic landscapes. The program used GAN, generative adversarial network, an AI model which provides the users with a “smart paintbrush.” It then detects what the user is roughly drawing, such a rock, hill, tree, etc., and then produces realistic textures to fill in the gaps in place of the drawing. Not only that, but it is also capable of taking reflections into account.
AI in the program can generate images almost instantly and requires almost no render time, which probably hints at what the software will already have a bank of stock images stored for all its operations. However, it still has its limitations. Right now Nvidia has only shown off a focused demo focusing on natural landscapes because that’s what it can handle pretty well. As for things such as buildings, cars, humans, etc., they are much more tricky to work with as those objects have “logic and structure to them that humans are sensitive to” (The Verge).
Though Nvidia has shown off an AI program which can generate realistic faces of people who aren’t real, merging that two software is a big undertaking. The software most likely won’t be released to the public as a full-fledged product, but instead, it will be used internally by industries such as architecture, urban planning, etc. It is meant to show off the capabilities of A.I., and within a few years, many incremental improvements will make it a powerful tool for these industries.
Perhaps one day this tool could be used as an add-on for graphic designers to fix minor changes and increase details in their renders. Alternatively, perhaps with enough innovation, this could be the future of art. Either way, this demo does a great job of showing something new that A.I. can do!