There is a new craze in town. Recently, designers have been typing prompts into a diffusion-based artificial intelligence (AI) platform and waiting for images of never-before-seen buildings, logos, products, and more to materialize within seconds. Platforms like Midjourney are built on data sets of billions of existing images scraped from the web. In this vast library, you will find pictures of buildings, birds, balloons, and beaks, so if a building in the shape of a bird with a beak made of balloons is something you are looking for, type it in and Midjourney will deliver. But beware—it’s addictive. In less than a month of using AI, I have created 11,515 images.
Midjourney (or DALL-E 2, Disco Diffusion, Imagen—there are many versions and more coming) is a text-to-image AI. In lay terms, it’s a web-based platform accessed through Discord (think: chat room) in which you type “/imagine” followed by a prompt, which is a description of what you would like the AI to create. Your imagination is the limit. For example: “/imagine a small house made of dinosaur fossils.” Hit return and the AI analyzes your prompt, searches through its database to find images to pair with your text, and then constructs four completely original images from a random pattern of dots. You have the option to upscale (add resolution) and vary any or all of the four images or run through a new iteration of your prompt. The AI fills in anything you left out of the prompt with elements related to the objects and parameters that would typically be associated with the content you provide. So, if you forget to include “doors” and “windows,” the AI will, in all likelihood, add them for you. On the other hand, if you want to exercise a little more control or replace doors with beehives, then add more detail to the prompt—color, material, entourage, mood, view, lighting, image aspect ratio, or even style—and run it again. Do this over and over (I have found the more upscaling and variation, the better) until you achieve a result that wows you. Or, if that particular thread is not doing it for you, type a few more words, and off you go with your next cocreation.
These images are designs coauthored (if you want to call it that) with AI, so I haven’t done the heavy lifting. We have designed houses on a lake, skyscrapers in Manhattan, hotel lobbies for a future when 3D printing and robotic fabrication are ubiquitous, housing blocks in the shape of letters, cities made of ingots, and even the background for the poster for our school’s fall 2022 lecture series. With this AI, there is room for conventional design, avant-garde speculative projects, and utopian (or dystopian) world-building. It’s hard to find the limits of its design capability.
I have no doubt that this will be a complete game changer, not only for architecture but for every creative discipline. AI is already deeply embedded in our lives (targeted marketing, self-driving cars, facial recognition), so it was only a matter of time before it found its way into architecture. Soon it will be in every office, every school, and every smartphone, and will play some role in the design process. The threshold of entry is minimal.
