If you’re not good at drawing, or photography, no problem. NVIDIA’s artificial intelligence gives you a hand.
Artificial intelligence is going to revolutionize our lives to unimaginable limits. A good proof is NVIDIA Canvas, a free artificial intelligence software that converts sketches into real photos, while you draw. All you have to do is make a couple of erasures, and the AI will convert them into a photographic image based on the sketch you’ve created. It is not a copy-and-paste of existing photos, but a completely new image that the AI invents, while you are drawing.
As we can see, with NVIDIA Canvas we don’t paint color strokes. We paint mountains, rocks, clouds, or seas.
We simply select the texture we want to use (trees, rocks, grass, etc), and when we paint the sketch on a freehand canvas, the AI converts it into a photograph with the same shape as the sketch.
NVIDIA Canvas uses GauGAN, a type of artificial intelligence called generative adversarial networks, where two opposing networks, called generator and discriminator, collaborate (or correct) each other.
The goal of the generator is to convert sketches into images. The discriminator knows that real ponds and lakes contain reflections and tells the generator to create a convincing imitation. To deliver realistic results, GauGAN has been trained on an NVIDIA DGX system using over 5 million images.
As we see in the video, GauGAN doesn’t just cut and paste pieces of trees, mountains, or lakes, but generates them based on the strokes, and adds realistic elements, such as the shadows of objects on each other, or how they reflect on the water.
The NVIDIA Canvas beta is available for free download, but unfortunately, it requires an NVIDIA Geforce RTX graphics card, which has been out of stock for months or shamefully overpriced, driven by cryptocurrency speculators, and the semiconductor crisis.