DALL-E 2 creates AI images of the future of San Francisco

“I keep waiting [Sutro Tower] to stalk down the hill and attack the Golden Gate Bridge,” quipped longtime San Francisco Chronicle columnist Herb Caen.

It’s easy for any San Francisco resident to imagine the scene, our city’s two most prominent symbols engaging in a Godzilla-style battle. But what happens when you ask a computer to actually create an image of the battle between landmarks? Or if you present the AI ​​with more complicated scenarios, for example by asking it for a picture of how climate change would affect the city?

Thanks to software from a new San Francisco-based artificial intelligence company, we now know. The DALL-E 2 image generation web app allows users to enter a text prompt and receive a series of six incredibly realistic images. Beta access users have produced everything from expansions to Michelangelo’s “Creation of Adam” to New York cartoons. And a publicly available competitor called the DALL-E Mini (and recently renamed Craiyon) has become a popular tool for meme generation.

The AI ​​imaging software DALL-E-2 created this visual interpretation of

AI imaging software DALL-E-2 created this visual interpretation of “The Salesforce Tower, but make it beautiful”.

Courtesy of OpenAI

The app is the latest creation of OpenAI, a non-profit Elon Musk and a host of other Silicon Valley heavyweights that launched in 2015 with $1 billion in funding. It shed its nonprofit status when it received another billion from Microsoft in 2019. So far, the tech company has developed technologies to teach machines how to debate, help robotic hands master movements, produce auto-generated journalism, and even create early iterations of a very flirty chatbot app. Given the inherent dangers of AI advances, the company’s releases are typically met with a mixture of astonishment and dismay, and OpenAI’s allegedly mysterious nature and changing financial status have raised suspicions from investigative journalists.

Dystopian deep-fake implications aside (the software has guard rails against showing real people’s faces and violence), the few users with access to DALL-E 2 are having a very good time with the software. It is currently in an invitation-only beta phase, with the stated goal of adding up to 1,000 users from their waitlist each week.

Although OpenAI would not grant SFGATE a beta account, we were able to provide a number of prompts. However, we learned that specificity is key, and without the ability to refine the input, the experience felt like telling a ghost you want to fly and getting a pair of mosquitoes in return. For example, the request to see the streets of San Francisco with gondolas instead of cars led to Venetian-style boats rather than high-flying ski lifts.

AI imaging software DALL-E 2 created this visual interpretation of what San Francisco would look like with gondolas, in the style of a painting by Norman Rockwell.

AI imaging software DALL-E 2 created this visual interpretation of what San Francisco would look like with gondolas, in the style of a painting by Norman Rockwell.

Courtesy of OpenAI

San Francisco concept artist Danielle Baskin (for the Google prank Spirit of Halloween and BART Basel Infamy) is an early adopter of DALL-E 2 and has spent hours playing with the software and exploring its limits. In April, she posted a Twitter thread showing what the Golden Gate Bridge would look like if a BART train ran over it, as well as a Sausalito full of high-rise buildings.


“My favorite thing is not necessarily doing anything with the results, I just love the feeling of exploring infinite combinations of images,” Baskin said. “It feels like stepping into a museum and I could walk down different wings and just see a specific part of history or an imaginary scenario.”

Although the software generates results in seconds, creating professional-quality photos isn’t instant. Baskin spent two hours refining their Golden Gate Bridge prompts. And some ideas, like showing what an additional living unit in a garage in San Francisco would look like, were just beyond the limits of the software’s conceptual imagination. Bike advocates will be sad to learn that building a bike-only version of the Bay Bridge without cars was also nearly impossible.

AI imaging software DALL-E-2 created this visual interpretation of a corgi flying over the Golden Gate Bridge in the style of a stained glass window.

AI imaging software DALL-E-2 created this visual interpretation of a corgi flying over the Golden Gate Bridge in the style of a stained glass window.

Courtesy of OpenAI

“From the outside, it might seem like the AI ​​is so good because you can just type in whatever you can think of and it spits out a beautiful image,” Baskin said. “But the first few were all messed up, the bridges mangled. I had to be really specific.”

When SFGATE provided DALL-E 2 with the question of what San Francisco would look like after climate change, the result was surprisingly conceptual, much like a publication would hire an illustrator to accompany an article. This way of using the software has understandably rocked the art world, with concerns that DALL-E 2 could put illustrators out of work.

AI imaging software DALL-E 2 created this visual interpretation of San Francisco after it suffered the effects of climate change.

AI imaging software DALL-E 2 created this visual interpretation of San Francisco after it suffered the effects of climate change.

Courtesy of OpenAI

With that in mind, Baskin has received several private requests from publications asking her to run command prompts for them. She found that in such situations, the media companies didn’t have the budget to commission anything and ended up using stock or creative commons images.

“Any kind of powerful technology kills jobs and creates new jobs,” she said, noting how useful the tool would be for illustrators looking for inspiration or helping refine their art to meet client needs.

The AI ​​imaging software DALL-E-2 created this visual interpretation of the cast of

AI imaging software DALL-E-2 created this visual interpretation of the Star Trek cast enjoying a day at Fisherman’s Wharf.


Courtesy of OpenAI

The AI ​​imaging software DALL-E-2 created this visual interpretation of the location of the treasure buried in Golden Gate Park as detailed in the book

AI imaging software DALL-E-2 created this visual interpretation of the location of the treasure buried in Golden Gate Park, as detailed in the book The Secret.


Courtesy of OpenAI

Three images created by AI software DALL-E-2 showing the cast of Star Trek at Fisherman’s Wharf and the location of the buried treasure in Golden Gate Park detailed in the book The Secret. Images courtesy of OpenAI.

In addition to exploring silly ways the cast of “Star Trek” would look like enjoying a day at Fisherman’s Wharf, Baskin sees potential for DALL-E to help expand people’s imaginations; perhaps it could even be used as a tool for activism. As the old saying goes, she has found that a picture is indeed worth a thousand words.

“A friend commented on DALL-E’s political power to show people very realistic alternate futures that they might otherwise be reluctant to envision,” Baskin said. “Many architectural renderings resemble blueprints or sketches rather than a strikingly realistic photograph.”

DALL-E 2 AI imaging software created this Renaissance painting-style visual interpretation of a perfect day in Dolores Park.

DALL-E 2 AI imaging software created this Renaissance painting-style visual interpretation of a perfect day in Dolores Park.

Courtesy of OpenAI

In terms of the prompts SFGATE provided about our fair city, DALL-E’s results were mixed. The aforementioned request for the gondola (in the style of a Norman Rockwell painting) and one asking for “a Renaissance painting of a perfect day in San Francisco’s Dolores Park” looked like paintings one would see on the wall of a college could see cafes. An admittedly dour request to show the city after an 8.5 magnitude earthquake looked like something out of a disaster movie trailer. Asking for help identifying the location of long-buried treasure in Golden Gate Park yielded no leads worth investigating, and our single attempt to replicate Baskin’s BART experiment failed miserably.

AI imaging software DALL-E 2 created this visual interpretation of downtown San Francisco after an 8.5 magnitude earthquake.

AI imaging software DALL-E 2 created this visual interpretation of downtown San Francisco after an 8.5 magnitude earthquake.

Courtesy of OpenAI

And as for Herb Caen’s old quote about the Sutro Tower attack on the Golden Gate Bridge? Well, sometimes a picture just can’t live up to a few well-chosen words.




AI imaging software DALL-E-2 created this Godzilla movie poster-style visual interpretation of the Sutro Tower attacking the Golden Gate Bridge.

AI imaging software DALL-E-2 created this Godzilla movie poster-style visual interpretation of the Sutro Tower attacking the Golden Gate Bridge.


Courtesy of OpenAI

AI imaging software DALL-E-2 created this Godzilla movie poster-style visual interpretation of the Sutro Tower attacking the Golden Gate Bridge.

AI imaging software DALL-E-2 created this Godzilla movie poster-style visual interpretation of the Sutro Tower attacking the Golden Gate Bridge.


Courtesy of OpenAI

AI imaging software DALL-E-2 created this Godzilla movie poster-style visual interpretation of the Sutro Tower attacking the Golden Gate Bridge.

AI imaging software DALL-E-2 created this Godzilla movie poster-style visual interpretation of the Sutro Tower attacking the Golden Gate Bridge.


Courtesy of OpenAI

AI imaging software DALL-E-2 created this Godzilla movie poster-style visual interpretation of the Sutro Tower attacking the Golden Gate Bridge.

AI imaging software DALL-E-2 created this Godzilla movie poster-style visual interpretation of the Sutro Tower attacking the Golden Gate Bridge.


Courtesy of OpenAI

AI imaging software DALL-E-2 created these Godzilla monster-style visual interpretations of the Sutro Tower attacking the Golden Gate Bridge. Image courtesy of DALL-E-2.

Leave a Comment

%d bloggers like this: