NVIDIA’s AI is perfecting generative art

NVIDIA (NVDA) recently launched version two of its “generative art” artificial intelligence (AI). This largely went unnoticed, but it’s a major breakthrough.  NVIDIA’s generative art AI is called GauGAN2. It’s named after the painter Paul Gauguin. And it can generate incredible, photorealistic images simply based on text and a rough sketch. Here’s a visual:

On the left is a sketch input to GauGAN2. The words desert, hills, and sun were included with the sketch. The AI took these inputs and produced the image on the right. It looks like a realistic depiction of a desert under a blazing sun.  And here’s the most impressive part – GauGAN2 produces these images from scratch based on its analysis of the sketch and the text input. 

That’s because NVIDIA used a supercomputer to train the AI on 10 million images as well as words that correspond with those images. It effectively learns from the training, contextualizes the data, and then produces an image.  This process is very similar to the way that you or I would remember how to draw an elephant standing on a tundra. 

The difference is, of course, that GauGAN2 can “ingest” and retain an exponentially larger amount of knowledge than we can. NVIDIA made GauGAN2 available to download to ANYONE with an NVIDIA RTX GPU. That makes this a fantastic tool for artists, graphic designers, and anyone working with digital imagery. 

These professionals or hobbyists can leverage the AI to both save time and improve their designs immediately.  That is… as long as they are running NVIDIA’s GPU. It’s a great way for NVIDIA to incentivize professionals to purchase the company’s hardware.