It all began with a friend’s text:
The idea of Don Quixote as an anthropomorphic donkey has enjoyed a long life in my head, because one can’t say “Don Qui-” without hearing “donkey.” The brain hooks these two disparate ideas together and creates an amusing image of a donkey + Don Quixote. I firmly believe this is a common phenomenon among English speakers, and I’ll die on this hill if I have to.
Lacking the artistic talent to execute even a passable sketch on my own, and unwilling to commission a painter to see such whimsy through, I kept it relegated to a passing notion – something my imagination would spontaneously provide for a brief chuckle now and then.
At last, feeling the power of God and anime on my side (along with the Wombo Dream app and WOMBOT), I had to see what I could do. I entered the following: a donkey sitting upright riding upon a horse, dutch windmill in the distance, gray sky.
If I had not been fiddling with the prompt for the better part of an hour, I would’ve loved to refine this result further by having the donkey hold a lance and a shield.
But my first prompt almost never gives me what I really want. The prompt I started with was: a donkey riding a horse with a windmill in the distance under a gray sky.
I didn’t save the images this initial prompt produced, but the variety of strangeness and dreamlike insanity amused me as much as it confounded me. After many iterations of borderline NSFW horse-donkey pairings and earth/sky inversions, I examined the prompt carefully and changed:
a donkey riding a horse with a windmill in the distance under a gray sky
to
a donkey riding upon a horse with a windmill in the distance under a gray sky
which gave me a tiny donkey on the back of a horse beneath a sleek, modern turbine. This was an improvement, but it still needed work. I tried being even more explicit:
a donkey sitting upright riding upon a horse, dutch windmill in the distance, gray sky.
This refinement rewarded me with the look I was going for. I was fascinated with the output, but I was also eager to know what was going on “behind the scenes” as the bot interpreted my request. This is my best guess: when I specified “a donkey sitting upright riding upon a horse” the AI was forced to improvise. It had likely never seen a donkey riding a horse before, so it filled in the gaps by putting a donkey head on a human body, which it definitely has seen sitting upright upon a horse. The addition of “dutch windmill” forced it to rule out modern windmills, along with other off-the-wall designs and abstractions. Finally, changing “under a gray sky” to simply “gray sky” avoided the AI inverting the landscape and sky. It seemed to me like the word “under” was leading the bot to think I wanted to see sky below the ground.
To recap, I started with this:
a donkey riding a horse with a windmill in the distance under a gray sky
and ended up with this:
a donkey sitting upright riding upon a horse, dutch windmill in the distance, gray sky.
If you’re interested in playing around with art prompts, my advice is to pay close attention to your prompt and its result. Figure out how to refine the prompt by comparing the output with your request. Add words that reinforce what you wish to see, subtract what isn’t working. Even if your work is whimsical, keep the bones of the description grounded. Another tip: try not to make the description too florid or verbose. An economy of words seems to pay dividends when working with these tools.