Nvidia Flaunts Art AI That Paints Realistic Pictures Just From Text You Give It
By Mikelle Leow, 23 Nov 2021
Video screenshots via Nvidia
Nvidia has been developing artificial intelligence art tools perfect for people who aren’t born painters. Earlier this year, the graphics processing giant launched ‘Canvas’, an AI painting app that transforms amateurish doodles into full backdrops. Now, you don’t even need to use your mouse or stylus—a few words are all it takes to channel your inner Claude Monet.
‘GauGAN’, the deep learning tool named after post-Impressionist artist Paul Gauguin that powers this technology, has been improved to generate beautiful photorealistic landscapes from scratch with a prompt of just three or four words. If you’re a wordsmith, adding another adjective or changing up the phrase will create a different look in real-time.
Enter “sunset at a rocky beach,” for example, and you’ll find an AI-created picture that shows just that. You can then replace “sunset” with “afternoon” or “rainy day” to switch up the scene.
Video screenshot via Nvidia
To fully tap into your creative abilities, the software can even go back to converting rough sketches, allowing the artist to mix text with imagery to bring their vision to life more quickly and easily.
AI of this kind usually has only one specialty, like inpainting or style transfer. This version, GauGAN2, however, packs multiple modalities of segmentation mapping, inpainting, and text-to-image generation into a single machine learning framework called a generative adversarial network (GAN).
“Rather than needing to draw out every element of an imagined scene, users can enter a brief phrase to quickly generate the key features and theme of an image, such as a snow-capped mountain range,” the company explains. “This starting point can then be customized with sketches to make a specific mountain taller or add a couple of trees in the foreground, or clouds in the sky.”
The image can keep evolving in tandem with the imagination, as “every word the user types into the text box adds more to the AI-created image.” You can get the AI to “paint” realistic pictures using vocabulary related to the weather, though it can also handle more fantasy-based prompts by, say, adding a second sun per your request.
The impressive image maker can be tested on Nvidia AI Demos.
#GauGAN2
— Merzmensch Kosmopol (@Merzmensch) November 22, 2021
"Blue Forest" pic.twitter.com/W6Wcwy7YUU
Text input only.
— Merzmensch Kosmopol (@Merzmensch) November 22, 2021
"A city full of buildings"#GauGAN2 pic.twitter.com/83BsMJN94g
Trying @NVIDIAAI #GAUGAN2 > input: "The Land of Certain Death" pic.twitter.com/Xmd0vTSw8D
— Giulio Bordonaro (@GiulioBX) November 22, 2021
"Universe"#GauGAN2 pic.twitter.com/9ZVdRImtAL
— Merzmensch Kosmopol (@Merzmensch) November 22, 2021
[via PetaPixel, video and images via Nvidia]