Make a Picture, Make a City

The next Street Stack walking tour will be in NYC this September. Details to come, but space will be more limited this time, and paid subscribers will get priority. It’s never too late to upgrade and help fund the large amount of time and effort I put into Street Stack.

When I was in grad school, I took a rudimentary class on visualizations. Below is an image I created for the unit meant to familiarize us with the Adobe Suite. I imagined a market made up of shade parasols that look like mushrooms. I stole the mushroom design from a real plaza in Europe whose name I’ve forgotten. I used the software to cut out images of these mushrooms and pasted them across a picture of a corner in Queens. I adjusted the colors, added some people and shadows, and got an image that’s passable enough to communicate the idea.

This market doesn’t make a lot of sense rationally, but the goal was to learn Adobe, not propose a good idea, and I had a disproportionate amount of fun making this collage, tapping the fossilized nugget of childlike enjoyment in my brain. More than that, by “rendering” the place, it somehow felt more real than if I’d just described it.

The next year I began working with an architect on a project in Saudi Arabia. This was my first exposure to actual architectural rendering, and although it happened at a much higher fidelity than my mushroom market, the principle was not all that different. Still images of 3D models were extracted and colored, and people and different set dressings were dropped in to create a composite. We spent a good deal of time in meetings thinking up scenes that suggested the human life we hoped would take root. Can you put in more children? The shopping should be busier. Shade the plaza with trees and put a couple picnicking below one. A guy I worked with actually had a framework based on Romantic painting he used to give feedback on renders — dramatic light source and people in the midst of some sort of emotional experience.

Towards the end of my time working with this architect, Midjourney, a generative AI tool that ran on Discord commands, launched. Midjourney created sketch-like images that were usually riddled with flaws if you looked closely. But it created images in seconds, and with iteration it did a passable enough job to populate proposals with secondary images. If we needed a quick picture of a landscape that wasn’t readily available on Google— show a desert oasis with mountains on the left side— or an image to punctuate a specific vignette — show a man in traditional Saudi dress enjoying coffee on a balcony with a metal railing— it was quicker than rendering it any other way. Some examples I stole from old documents below:

These images look quaint compared to the high-fidelity AI visualizations flooding our feeds just a few years later, fooling boomers (and… me) into thinking they’re real. The debate around AI images is often framed as truth vs fiction, but the exercise of rendering is speculative anyway; we are basically creating fantasy. Here is an image we created the “old-fashioned” way, which is to say with Photoshop. It’s all obviously fake, and I don’t think there is a moral issue creating this type of picture with AI now instead of Adobe.

While I’m skeptical about AI’s usefulness, and I’m fairly certain of its harmfulness in many ways, I’m excited to see how it pans out for planning in this one regard. Visual literacy and rendering ability are not required tools for planners. The common problems of planning- where to put new housing, creating mobility networks, stimulating economic development- are aesthetically agnostic. But communication is a vital element of planning. Planning ideas need to be seductive to succeed, or at the very least they need to be legible to their stakeholders. Language plays a part. “Garden-city” is an evocative term; “eyes on the street” is a snappy way for Jane Jacobs to sum up her complex ideas. But pictures work even better. Many of the foundational figures in planning were talented image makers. The hand-inked maps and oil-painted cityscapes of Daniel Burnham or Olmsted fired their ideas into the popular imagination.

On the flip side, many architects get a lot of attention for their dubious planning ideas off the strength of their images. I would never want to live in the vast, car-dependent landscape of Le Corbusier or Frank Lloyd Wright, but they’re fantastic drawings.

More than just pretty drawings, these images inspired real projects. One lesson I learned from my brief time in architecture is that cool-looking shit usually wins. Visualizations don’t just represent ideas; they influence our reality.

Recently, I got early access to visualization features being launched within Google Earth. By aiming the camera at a point on the map and writing some rudimentary prompts, it was possible to say, fill in the East River with housing or turn Manhattan into a forest. Reading into the backend on how these images are made, Google Earth is going beyond simply redrawing the image according to the prompt; it’s pulling in information from multiple sources about the subject you’re prompting it on. One telltale sign is that without me asking, Google Earth used an actual NYC Ferry in the visualization even though I hadn’t specified this. Here is a park I asked the system to create on an empty lot in Queens. Despite a few quirks (the scale of the ferry is off) it’s not bad.

Google has apparently stunted the rollout of these features after people used them to create maps showing things like smoldering craters where the White House should be… or, on the other side of the spectrum… this:

The above example is an obvious joke. More concerning was the extremely realistic images of refugee camps and Iranian Nuclear plants that the writer Henk Van Ess was able to create with the tool.

That AI is accelerating misinformation and slop seems like a given. Maps created to understand the world as it is today need to be trustworthy. But past the junk, in the speculative future of planning, what could it mean if image creation is freed from artistic or technical ability? The ability to plan is largely restricted to large private developers, governments, architecture firms, and consultants. These entities have the technical know-how to create plans and the power to get them implemented. Can you imagine a world where you, with careful prompting and some research, could bring forward an idea for an empty lot or your neighborhood park that looked almost as good as what an architect hired by a developer could? This will never fully replace actual expertise, but it could drive what Paul Davidoff called a “plurality of plans,” an environment where planning isn’t top-down but stakeholders can arrive at a planning process on more equal footing to select from a range of ideas, not just react to one proposal handed down from on high. In this VERY SPECIFIC RESPECT, I find myself to be a surprising techno-optimist.