I Asked an AI Robot to Imagine Iconic Bay Area Scenes
Redwood trees in sunlight. (Photo credit: Thomas Smith via NVIDIA GauGAN)
NVIDIA is known for creating mind-blowing AI technologies. The Silicon Valley company makes graphics cards for gaming and cryptocurrency mining, but they also put those cards to use performing all kinds of AI wizardry.
The company was one of the originators of Generative Adversarial Networks, the technology behind Internet sensation ThisPersonDoesNotExist.com as well as numerous more practical generative AI systems.
Now NVIDIA has shifted their AI focus to something else: using deep learning networks to imagine places and scenes that don’t actually exist and to create them in a computer in photorealistic detail. The company’s new system for doing this is called GauGAN, and it’s spooky and amazing. You give it a text prompt like “a tranquil forest” and it creates a fully-realistic scene based on your prompt. You can play with it yourself on NVIDIA’s website.
I decided to put the system through its paces by asking it to create a variety of iconic Bay Area scenes. As a photographer, I love the natural beauty of the Bay Area, and the diversity of different environments and microclimates it includes. You can go from crashing waves on cliffs in Pacifica to the rolling hills of the East Bay, to rows of perfect vineyards blanketing the landscape in Napa or Sonoma in a single day if you want.
What better way to test out NVIDIA’s new system than to ask it to generate some of the iconic scenes we encounter here by the Bay? Here are the prompts I provided and the results — the beautiful, the terrible, and the downright creepy.
‘A Tranquil Forest of Coast Redwood Trees’
Photo credit: Thomas Smith via NVIDIA GauGAN
When many people think of the Bay Area and the Northern California coast, they think of redwood forests like Muir Woods. The tranquil floor of a big redwood forest is one of the most peaceful places in the world, and one of the most beautiful things about the Bay Area.
GauGAN does an admirable job of capturing the feel of a redwood forest, both in this picture which reminds me of the woods near Tilden in Berkeley and in the photo at the top of this article.
‘Fog Over a Body of Water’
Photo credit: Thomas Smith via NVIDIA GauGAN
For all my best efforts, I couldn’t get GauGAN to create Karla the Fog. But the system did a cool job of drawing this foggy sunset over a pier. It kind of reminds me of standing near the water in Emeryville and looking towards Oakland over the commercial piers there.
Does it matter that fog rarely persists until the evening in the Bay Area, since the sun usually burns it off? Maybe. Still, this is a cool sunset shot.
‘Ocean Waves on Pacific Coast’
Photo credit: Thomas Smith via NVIDIA GauGAN
I love this one — it really captures the majesty and power of the Pacific Ocean’s surf on a windy day. These waves are too aggressive to be Ocean Beach, but I could totally see a scene like this at Mavericks near Half Moon Bay.
‘Fishing Boats in Port’
Photo credit: Thomas Smith via NVIDIA GauGAN
Granted, this one isn’t totally photo-realistic. What are those weird house-like things floating in water in the foreground? Why are there so many of them? Maybe Google is testing another set of mysterious barges?
What I like about this one, though, is that it evokes the feel of an actual place — the industrial port right at the base of the Richmond-San Rafael bridge on the Richmond side. A big dock accommodates oil tankers, and various buoys float in the foreground in that actual place. This photo isn’t perfect, but it captures that industrial feel and the general layout of the place.
‘Cars in Rush Hour Traffic’
Photo credit: Thomas Smith via NVIDIA GauGAN
This one is one of my favorites. It’s not photorealistic, but I think that actually makes it better. It doesn’t feel like GauGAN is trying to create a real depiction of a rush house scene — it feels like the system is trying to make abstract art.
The converging lanes here feel like they capture the chaos of rush hour by the Bay. I love the drab, gray colors interrupted by the vibrant reds of distant brake lights, and the weird little stacked rectangles that evoke the idea of the small living spaces many Bay Area drivers are returning to.
I know this is the hallucination of a computer, but I can’t help feeling it captures something of the essence of a Bay Area commuting life.
As AI advances, systems like GauGAN will become even better at creating realistic scenes. Mutant llamas will presumably disappear. The system will likely even be able to create complex scenes, like a cityscape or a scene of a cablecar cresting a hill. In the meantime, we can enjoy this baby step — and the strange artistic mistakes it facilitates.
This article originally appeared in the Bold Italic