Conjure Up A Memory
What Happens When You Challenge an A.I. Image Generator to Match Your Vision
Like many of you, I can’t draw. So, the promise of artificial intelligence creating an illustration based on a description I type in seems as exhilarating as music illiterates being able to lay down scores by simply humming. Creativity unleashed!
In this inaugural column, I ask an image generating program powered by A.I. to create a set of pictures so I can choose the one that best expresses what I have in mind. Typically, my request is based on a real-life story, so I am admittedly prejudiced about what I envision. I try to tailor the instructions accordingly. In the end, the winning digital creation is determined by what my analog brain imagined all along.
A.I. Challenge #1:
Man Attempts to Bribe Younger Man with a Pickle
In this first test, I ask the remote server-based program, DALL-E, to create an image of an older man trying to bribe a younger man with a pickle.
To get a sense of an A.I. image generator’s innate style, I didn’t ask it to create an image in the style of Dali or Rockwell, or anyone.
Since this incident really happened, it’s important to understand that I want to create the image with a certain spin. DALL-E spit up four choices in less than 30 seconds. Not all of them fit the bill.
Back Story: as a young reporter for a daily newspaper, I was handed a press release from the New Jersey Department of Health citing such violations as rat droppings and fly infestation at a pickle plant in Paterson. My job was to get a comment from the company and write a story. But the man responding to my call, who claimed to be the boss, tried to stop me from publishing anything by offering me free pickles. Just come around to the back door, he implored. Though I hated the flavor of cucumbers, I did like full sours. Still, trying to change my mind by bribing me with food produced under suspect circumstances wasn’t exactly a winning strategy. I turned down the offer and ran with the story—leaving out the part of the failed bribe.
Here are the images the program created in order of worst to best:
The interplay is all wrong. The older man appears to be smiling as he hands the pickle to the younger man, also grinning. As for the pickle, it appears to be half sour, meaning too much cucumber taste and not enough briny acidity. The attitude being expressed is half-baked. What’s the point? The older man should appear threatening and the younger one somewhat scared. Score: D-.
The older man should be looming over the younger man. Not intimidating at all. Besides, what’s with that tan? Looks like instead of working inside a processing plant, he’s just dropped in from a week at the Jersey Shore. D+
The older man exudes the charm of the late Wilford Brimley, the Quaker Oats’ spokesman. His appearance is way too folksy for a transaction that’s supposed to be an underhanded bribe. Not convincing. C-
Finally, in this image the two men appear to be in a heated discussion. No one is smiling. In fact, the older man appears earnest, even desperate, realizing that if word gets out about the unclean conditions at the pickle plant, he and the factory could lose a ton of money. He seems to be grasping the pickle so hard that the color has drained from the top of it. The man’s outstretched hand seems to be saying. “Give me a break, buddy.” But the reporter is just as sober. He’s not taking the pickle, and his hand is in the form of a clenched fist.
The color of their clothing works: blood-magenta for the boss man; hospital green for the reporter. What I’d change though is that the younger man should not be wearing a cutoff T-shirt. As a serious reporter, he should be in business attire. I should have included fashion notes in my instructions. Also, the older man should be pointing the pickle toward the reporter like a gun—not prayerfully toward the sky.
The image, nevertheless, comes closest to my vision, capturing the menace of the moment. I award it a B.
A.I. Challenge #2:
Isn’t It Romantic? (It being A.I.)
What’s more romantic than a candle-lit dinner in a nice restaurant? Well, if you had your heart set on atmosphere, better think LED. More restaurants are saying no to wax by installing digitally operated table lamps instead. The battery at the base of the light emitting diode, often atop a long-thin stem, charges through a USB port. When patrons aren’t there, the restaurant regularly plugs the high-tech lamp into an outlet for recharging. Unlike a candle, the LED is never consumed and lasts for years.
But a patron might rightly ask: where’s the warmth? In fact, there’s none. Hold your finger to the light, and it’s cold as ice. Where’s the erratic twinkle as reflected from the candle in your partner’s eyes? Forget about it!
Of course, there’s something to be said for the banishment of table candles. First, noxious fumes go away, especially the one left by a match. No one wants the scent of sulfur fouling the food, or the lingering smoke that causes your date’s contacts to tear up.
This is where the refrain, “You can’t fight progress,” resounds. The phrase is also used to answer patrons who complain about having to point their phone at a Rorschach test just to see a menu.
Maybe romantics should be grateful that the restaurant candle is gone. Who needs another fire hazard, especially when reading a combustible newspaper while waiting for your date to arrive? Oh, wait! Hardly anyone carries a newspaper with them anymore. They’re too busy checking their screen phones. In fact, the post print era could have been the golden age of table candles. Another missed opportunity!
Back Story: I admit that I’ve sometimes been oblivious and inadvertently set a newspaper on fire. It was in a long, narrow Italian bistro on Park Avenue South in New York City some years ago. The paper was ruined, but the bread sticks were fabulous, and my date became my wife.
Now that you know the truth, I recently tried an auto image generator powered entirely by A.I., a.k.a. artificial intelligence. So, into DALL-E I typed: “Man is seated in a restaurant. He reads a newspaper about to be set on fire from table candle.”
DALL-E returned four images in under 30 seconds. All of them conveyed the necessary ingredient of absentmindedness on the part of the diner. None of the men were aware of pending doom. The diner was not injured, though I’ll leave it to your imagination whether the guy’s hand gets singed.
I’ll rank the four images from least- to most-favorite according to what I perceive as the best in capturing my true-life situation.
You know when at the audition the actor tries too hard? This how I feel about this picture. The guy has been rendered hiding, head down so deeply into the news that we have no idea who he is—aside from the receding hairline, that is. Also, the foreground is way out of focus, making the presence of candles nothing but a blur. Score: D.
I don’t get the sense of a coming conflagration here. It seems the guy can pull away at any moment, saving the paper and saving face. C
I found this image the most artistic because of the symmetry. The fire is reflecting off the newspaper’s flag, crackling from the fireplace (a nice touch in winter) and emanating from the table candle itself. True, I was put off by the guy’s nonchalance and his open-buttoned shirt that is about to be dissed by marinara sauce when the meal comes. B
This spot art clenched my heart. It captures the tipping point between normalcy and lunacy. If the paper hasn’t already ignited, you know it will in microseconds. Also, I like the reflection in the man’s glasses, and he is wearing a suit and tie, looking his best to impress his date. B+
Conclusion: An A.I. image generator can be romantic, though in a Hallmark Card kind of way. I do wonder, though, if I hadn’t expressly asked DALL-E for a table candle, it might have conjured up an LED by default. Still, for an incompetent artist like me, all I can say is Hallelujah!
Moral of the story: Always pay attention to your surroundings. Also, if your date arrives just in time to see your newspaper catch fire, he or she should head the other way.
A.I. Challenge #3:
Rabbi Sells Camera to Customer
In this test to generate an image, I ask DALL-E to first show me “a rabbi in a camera store selling a professional camera to a customer standing in front of the counter.” After the first set of images are created, I revise my instructions to state: “A rabbi behind a store counter sells a camera with a long lens to a customer in a flowery shirt and shorts.”
Back story: I once worked in the back office of a camera store where salesmen behind the counter prayed regularly and dressed like rabbis. Despite the anomaly of spiritual seekers pushing electronics, there were plenty of satisfied customers.
I’ve chosen to highlight four of the eight images that were generated leading up to my favorite.
First, I don’t buy for a minute that the salesman in this picture is a rabbi. Amish, maybe. And the customer is much too young to be sporting a beard. Also, what exactly is he buying? It looks more like mittens with a disembodied thumb all rolled up in a polyvinyl bag. But a camera? F+
I do like the Island vibe, but the Hawaiian shirt worn by the salesman goes too far. No rabbi I know dresses this way. Besides, I only requested that the customer be decked out in a flowery shirt and shorts. D-
I like that the salesman is friendly, but is that grin trustworthy? His full red beard reminds me of a rusty but reliable Brillo soap pad. He’s pitching the customer on the merits of the merchandise, though the product looks more like a speedlight for the top of a camera than a camera itself. We only see the customer from the back, so he can pretty much stand in as an every man. The image is just serviceable. C+
This one takes the cake in the sense that we have a serious-looking salesman holding a long lens (as instructed) out to a customer who seems genuinely interested in either the product or the salesman. Not sure why she’s wearing a bonnet, but maybe hats are back. Considering how prim the customer looks, I’m surprised that she’s not wearing a ring on her finger, but maybe I’m being out of line. The hair color on both is nearly identical, making me wonder if the two are related. In any case, this image is the one that presents an over-the-counter transaction I can live with. B+
A.I. image generation at this stage will never be confused with a Rembrandt or Van Gogh, but it does allow an inartistic person like me to add pictures to his prose.