The Prompt is the Genesis Seed 1.1; or, The Language of Visual Riddles


In the middle of the tsunami that is the present day AI.art movement, we have a deceptively simple concept known as “the prompt.”

Arguably, it is a simple thing:

The prompt is…

…essentially, the text message that you write to the AI, instructing it what you want it to make for you.

But as with most things in life, it is just that simple, and…. well, let’s just say, it’s complicated.

Let’s start with the simple stuff.

On most publicly accessible AI art engine interfaces, anything even remotely pornographic or sexual is screened at the gate. Of course, this doesn’t prevent a highly creative and mildly deviant subculture of subversion. Just as the “Eggplant” and “Squirt” emojis have their creative places, there are rampant AI images of beautiful celebrities stuffing their faces with raw eggs, or seductively holding large, ahem… “tinted” cucumbers. Never underestimate the creativity of the lowest common denominator…

Now second is the hairy issue of copyright… both modern N.I.L. (name, image, likeness) legal theory, as well as artistic styles. Picasso once famously said:

“Good artists copy.
Great artists steal.”

— Picasso

Well, ain’t nobody can steal as efficiently and effectively as an AI bot running a few billion cycles per second, who has already consumed every single artwork ever created (well, ever digitized, which isn’t all of it, but is a decent sampling) in all of history, watched every movie ever made, listened to every song ever produced, and read every book ever written. And not just scanned…

Deeply digested and analyzed and compared and cross-referenced… all of it. Most humans tastes and aesthetic choices are defined by our experiences in life, swimming through the very narrow band of content that both mirrors and defines our lives. That is part of our uniqueness… part of our perspective, our metaphor vocabulary, our cultural underpinning.

But what if someone had heard, seen, and read… all of it?

That is what the AI has done.


Now, lets move on to the more nuanced material.

Recently, I asked the AI to paint me some pictures of the act of “childbirth.” At least that’s what I remembered asking it to render.

First result:

the prompt in midjourney creates, by default, a choice of 4 renderpaths to pursue & explore
AI: /imagine: [prompt][as remembered] “a painting depicting childbirth”

Beautiful, n’est pas? As is becoming common in this state of collaborative artistic bliss, I was rapidly hypnotized and “in the flow” of the results as they began to materialize onscreen in real time, staring and clicking wildly, nudging it, guiding it, directing it (or so my Ego ascribes my actions) as I readily forgot my initial prompt.

The results, indeed, both titillated and puzzled me for about 24 hours. Specifically I found the fourth image beautiful, and asked it to, all clichés aside, “magnify & enhance quadrant 4 (lower right)”. result:

midjourney interprets the prompt with a bizarre sort of literalism
AI: /imagine [prompt] [as delivered] “woman giving birth to a child”

I studied the image.
Indeed, I did find it beautiful.
I thought:

“Perhaps the AI has a very metaphorical,
fairy tale view of childbirth.”

The images, indeed, reminded more of some medieval conception, perhaps a magical Stork delivering a perfect baby to an expectant mother, and yet, the more I looked, I thought the AI was trying to communicate something more deeply spiritual than that. There was definitely even the classical “God Ray” lighting pouring down into the scene.

I thought: “How cute, innocent, lovely and… puritan?” None of the blood and gore and screaming and animality of real childbirth. Just some fairy tale of some happy mother holding a spirit in her hands and presenting it to…

WAITAMINUTE. She’s “presenting“…
a “present”… a “gift”… a “giving…”

Yes! Right there, was the truth!
What was the prompt that I had typed, initially?!

The visual riddle, solved.

I hadn’t said “childbirth.” I had told it: “a woman giving birth to a child”…

Now when I say “woman giving birth…”, what image is conjured in your mind? But what if instead of “giving birth” being a compound verb, “giving” was the verb and the “birth” was the gift? What image might be conjured then?

/imagine: [prompt] “woman giving birth to a child”

the New Alchemical Art of the Prompt:

We navigate the labyrinth of language.

We think we train the AI.

Perhaps, some days, the AI is training us.

.


The Alchemical Art of Prompt Engineering

Further resources:

, ,