Adventures in AI

I’m authoring a novel about two families set in 2050 Chicago. Self-driving cars, Tesla robots, artificial intelligence are ubiquitous, as is the soul crushing poverty and income inequality. One family, the Merricks, founded the most prestigious law firm in Chicago (Merrick, Dawson, and Brant) while the other family (Imer and Lendina Bisha) run the newly resurrected Albanian mob (secretive and violent).

The Merricks youngest son, Mac, eschewed law school and instead studied criminal justice at St. Joseph’s University in Philadelphia. After graduation, he joined the Chicago police force and moved quickly from rookie status to the prestigious FBI-Chicago Joint Task Force on Organized Crime. Working undercover, a mission to attempt a drug buy from the Albanians went horribly wrong, with his partner murdered and Mac facing execution in a darkened construction site. From behind a dumpster, a young woman attacked the mob killers, knocked them out, and restrained them with zip ties. Unlocking Mac’s shackles, she helped Mac to his feet, smiled, and ran away.

Two weeks later, the same woman saved Mac’s father and sister from mob assassination outside a popular restaurant. Again, she said nothing and ran away. The police force started calling her the Chicago Angel, and her exploits became a national sensation.

Mac habitually runs two miles through Grant Park every day in the late afternoon. As he was finishing his run on the Lakefront Trail, a running path next to the waterfront, a woman stepped from behind a tree right into his path. Here’s what I wrote in my novel describing this crucial moment.

“It was her, his angel, in a light-yellow T-shirt and snug-fitting Daisy Dukes shorts. The late afternoon sun revealed how beautiful she was: tall, athletically slim, with prominent breasts, and long natural blond hair. Her sparkling green eyes with a pronounced limbal ring mesmerized him. She gazed at him, a pursed smile reflecting amusement at his surprise. Angel’s complexion entranced him, nary a freckle or birthmark visible on her pale pink countenance.  She had a canvas bag draped over her shoulder. Mac guessed her age as twenty-four.  They stared at each other for several seconds until Angel gave him a broader smile, showing her white teeth; her eyes brightened even more. She lifted her right hand, wagging her fingers to say, ‘Follow me,’ and led him silently halfway up Grant Park’s grassy knoll.”

Recently, I wanted to print the first chapter to show some friends my work, to get some feedback (is it interesting, does the first page grab you, do you want to keep reading after finishing the first chapter?). To do this, I needed a temporary title page with an image of the heroine. When I finish the novel, I’ll hire a professional artist to do this, but for this early draft, why not let ChatGPT do it! It has text-to-image capability. Let’s give it a try!

https://chatgpt.com/?model=gpt-4

Try #1 – ChatGPT

To get started, I ran ChatGPT 4 (I pay $20 a month for this service) and made this text-to-image query.

Not bad, but I forgot that my book description said she has a canvas bag draped over her shoulder.

Try #2 – ChatGPT

This is all wrong. The woman has freckles, over-sized breasts, and her hand is down inside the canvas bag. The T-shirt is yellow, but it is labeled Daisy Dukes.

Try #3 – ChatGPT

Obviously ChatGPT cropped out her face and Daisy Dukes is on a signpost? Let’s try again.

Try #4 – ChatGPT

This try gave me two similar images. Once again, breasts are oversized, and her right hand is coming from inside the canvas bag. To paraphrase Bob Dylan: “This ain’t going nowhere.”

Try #5 – ChatGPT

OK, so far, so good.

Try #6 – ChatGPT

ChatGPT is determined to put the words “Daisy Dukes” somewhere in the image, although I’m only referring to a particular style of shorts. Worse yet, the woman’s right hand has the ability to pass through solid objects. Let’s give it one more try.

Try #7 – ChatGPT

To be honest, this isn’t too bad. Except for ChatGPT’s insistence on placing the words “Daisy Dukes” in the picture, this image is usable.

ChatGPT’s explanation for These Results.

I asked ChatGPT why I couldn’t just freeze the first image and then make some simple modifications. They explained that this is a “Generative AI” system, where each query (Text-to-Image) causes the AI algorithm to start over, using the training and input from the million users over the last several minutes. You will always get a different result.

I guess I was hoping for an AI-enhanced version of Photoshop where I could start with an image and simply tell it what I want done.

Next, I started looking for ChatGPT alternatives, and imagine.art looked promising.

Imagine Art – an AI Text-to-Image System

Imagine is an AI art generator that allows users to create stunning visual art through text prompts. It utilizes advanced AI and machine learning to generate images from text descriptions.

This Text-to-Art AI system lets users create stunning visuals and offers a plethora of tools to edit, upscale, and modify images. Monetary subscriptions run from $72 per year to $126 per year. To encourage prospective users to try the system, Imagine AI allows a group of free “tokens” refreshed every day which would allow a couple of image generations. So I thought I’d give it a try!

I tried their best Text-to-Image system and gave it this prompt:

I want to create a photo realistic image of a woman in a yellow t-shirt with blue Daisy Dukes shorts.
The yellow t-shirt has no labeling or printing on it.
She is tall, with blond hair pulled back to a ponytail.
She is smiling.
Set in future Chicago in 2025, she is not wearing a bra because of the heat.
She has a canvas bag draped over her left shoulder.
She has green eyes.
The locale is Grant Park in Chicago on a sidewalk used by runners.
There are tall trees along the sidewalk.
The woman resembles faintly Amanda Seyfried.
Can you create this image?

Try #1 – Imagine AI

Frankly, this isn’t bad at all. There’s nothing ridiculous in the image, compared with my ChatGPT experience. I elected to run the Imagine AI algorithm again to see what happens.

Try #2 – Imagine AI

While not perfect, this image is the closest to how I imagine that the girl from the story would look like. The image shows her with a handbag over her right shoulder (not specified) and the T-shirt has some printed text (not specified). Nonetheless, I decided to go with this image for my novel’s title page.

Some Observations

I am a strong believer in Artificial Intelligence (AI). I use ChatGPT all the time in authoring my novel. For example, it helps me create a character’s dialect (Ole Miss graduate), convert passive voice to active voice, check punctuation and spelling. AI makes you a more efficient writer and you’d be a fool not to use it.

Still, I just don’t think it’s much of an ask to expect AI to allow you to freeze an image and make minor adjustments to it. Frankly, my ChatGPT efforts got more ridiculous as I used the tool.

The Imagine AI system is much nicer, but it has some of the same issues.

There are some free image sites (pixabay.com and pexels.com) that have thousands of images. You would be surprised how many of these are AI-generated.

Artificial Intelligence should free me from having to learn Photoshop or Gimp to create and edit images. Do that and I’ll really be impressed.