Gawk at these DALL-E renderings of DC streets freed from cars
What would Georgetown’s M Street look like if, instead of six lanes of car-dominated asphalt, it were covered by brick paths, flower beds, friendly fountains, and zero cars at all? It turns out artificial intelligence can show us.
Using an AI called DALL-E, artist Zach Katz has worked out a method to turn photos of real-life car-dominated streets into pedestrianized nirvanas, with amazingly cool results. Observe:
7th Street (Washington, D.C.) pic.twitter.com/81MzmVm5nk
— AI-generated street transformations (@betterstreetsai) July 28, 2022
Logan Circle (Washington, D.C.) pic.twitter.com/zqyHsGetff
— AI-generated street transformations (@betterstreetsai) July 27, 2022
The renderings aren’t as precise as the kind of expensive and time-consuming creations typically included in government plans. Zach can’t specify that a bike lane be exactly 10 feet wide, or that a flower pot sit exactly six inches from the sidewalk. But he can make these renderings in only a few minutes each, needing only a good base photo, a thoughtful description, and some AI computing power.
It’s a remarkable step forward in being able to visualize what a street transformation might look like, at least in general.
Zach’s Twitter account for the project, @betterstreetsai, rocketed from zero to over 10,000 followers in the span of ten days, as he produced image after image showing street transformations all over North America.
US-19, “the deadliest road in America” (New Port Richey, Florida) pic.twitter.com/dZRsaALn83
— AI-generated street transformations (@betterstreetsai) July 27, 2022
Thompson Street (Greenwich Village, NYC) pic.twitter.com/VJy102dA7u
— AI-generated street transformations (@betterstreetsai) July 25, 2022
You can play with the free version
To modify an existing photo of a real street, Zach is paying for the professional version of DALL-E. At the moment the pro version is not accepting new users, but you can send Zach a request or a donation on his buymeacoffee.com page.
If you want to play yourself, there’s a free version of the AI, DALL-E Mini, that’s nearly as powerful. It doesn’t allow you to set a base photo of a real street, but you can use it to produce AI-rendered images based on written prompts. Type in something like “tramway in a street” and the AI creates a series of images of that thing:
Add more detail to your prompt, and get more detailed images:
What can you come up with?