I was archiving my downloads folder the other day, and found some images I generated with a text to image AI–late one evening in November of 2022. Initially I was disappointed, after hearing so much buzz about this advance in computing, as they missed the prompt largely. They did not meet my expectations at the time. Upon review, I see these images a little differently. Below, I will share them with you, as well as an effort to analyze them. Scroll down, I will describe what I see, from an architectural historian’s perspective. I will also draft up some features of this output.
“architectural section perspective starting underwater and ending on land”
That was the prompt I would like to discuss here, that I used in my first AI Text to Image experiment in November of 2022.
What I would like you to look for is the fulfillment of the prompt, the alignment ,or lack thereof, in the images to the text. Are these architectural section perspectives? Do they start underwater and end on land? I wanted to prompt the AI with something that would be challenging for any draftsperson. Architectural sections have certain conventions and architectural section perspectives are even more challenging. None of those are apparent. What is shown is a background. Perhaps one that an actual architectural section perspective could be drawn over.
Rethinking this project now I could improve the prompt’s logic with the following:
“architectural section perspective of a building that starts underwater and ends on land”
Granted, making a prompt crystal clear in terms of logic would likely yield repeatable and expectable results, so I gave it something harder to visualize and harder to imagine. How often do buildings start underwater and end on land? What ways could that be accomplished? There are many.
For example, in this image the land portion, from a distance looks like high end residential waterfront properties, and the below the surface portion resembles a gray, murky, industrial disaster.
Next I am going to zoom in to some areas of interest.
This first sample form the images, of the “ends on land” above ground, reveals a kind of unrecognizable lush yet inarticulate landscape. The buildings along the waterfront, are patchy and vaguely architectural. The plants are generic and green but nonspecific.
This second sample from the “starts underwater” portion below the surface shows a very unusual spindly series of buildings along the diagonal, resembling attenuated water towers, or bygone factories, or cranes. These structures are elevated on spindly pilings, connected by catwalks and spiral ramps. Only precedent I can think of is Yakov Chernikov’s Architectural Fantasias.
This is the first time I took the time to look closely at these generated images, and my first impulse was to draft one up. So I did.
Again, I was initially disappointed with the results of these text to image generations as they did not fulfil the prompt in my view. I made five or so other prompts on that November 2022 evening. Now that I look again, my interest is renewed. May write more about those first impressions, may draft more or generate more, not sure. Thank you for reading.
drdrr.work 