Microsoft Paint was a fun, creative outlet for many bored early Windows users. Even today, the simple software offers scribbling abilities with a blank canvas to digital artists who have nothing better to do.
Now, an AI program from NVIDIA can turn those simple doodles into a work of art with the click of a button.
With the new GauGAN system, users can create stunning landscape canvases in just a few seconds. Tech like this can turn anyone into Bob Ross as long as they have access to a mouse. The program also demonstrates the mind-bending power of AI-driven image rendering.
In most paint programs, users have the option of selecting a color and a brush type. GauGAN’s interface is a bit different. Rather than picking a color, users choose a label like “tree,” “stone,” or “water.”
Next, all the artist has to do is lay down a few strokes on the “canvas.” The AI underneath the hood of GauGAN takes care of the rest. For example, depending on its size, a swath of blue will be rendered as a lake, pond, or ocean. Or, drawing a thin line with a circle atop it will display a mighty oak.
Once the picture is complete, users can customize the look of their landscape. In a software demonstration, developers created a rocky ocean landscape and applied different styles to it like stormy, sunny, abstract, and more.
Furthermore, artists of multiple skill levels can upload a sample of their original work, and the program’s AI will render the image in their unique style.
Better Than Any Scribblenauts
Video game enthusiasts may remember the “Scribblenauts” games from systems like the Nintendo DS. GauGAN primarily works the same way by taking simple doodles and transforming them into a more detailed form. The playful demonstration shows some of the future real-world applications for AI rendering, including making video games.
GauGAN can depict an environment with incredible detail with just a few strokes of the digital brush. More impressively, the system can recognize elements that would alter the final picture. For example, painting a tree near a lake displays its reflection on the surface of the water automatically.
According to VP of Applied Deep Learning Research for NVIDIA, Bryan Catanzaro, “The neural network is able to fill in all of the detail and texture, and the reflections, shadows, and colors, based on what it has learned about real images.”
This sort of tech could be incredibly useful for video games with a large map and an open world. Developers could program a few sketches into the game, and an AI engine would render the world based on the season, the player’s point of view, or even the time of day.
If used properly, such a system would significantly reduce the amount of work and processing power needed to create today’s game environments. Currently, this relatively simple program is just in a beta demo phase. However, don’t be surprised if a similar application is rendering fan favorite game graphics in the near future.