Is Human Art REALLY Inherently Better than AI Art?

David Recine
May 21
9 min read

Three weeks ago, I made a post where I pitted my own stick figure drawing skills against AI. My prognosis? Don't use AI-- any human drawn stick figure will look better and take less time to create.

Elevate AI Coaching's CEO, Chris Lele, graciously allowed me to publish that article on our blog, even though he had concerns that my argument is a bit... well... rigged. How would Chris say it was rigged? Well, he (and others) have accused me of being a "professional stick figure artist." How do I plead? I will enter my plea in Doctor Doofenshmirtz Nickel Meme format:

In a two-panel cartoon image, Doctor Doffenshmirtz, a character from the Disney TV show "Phineas and Ferb," stands in front of robots. In the top panel, he humorously states to a sock puppet that he's earned $3,000 drawing stick figures over the past decade. In the bottom panel, he acknowledges the small amount but finds it odd that it happened at all.

Guilty as charged! And given how rare it is to get paid to draw stick figures, perhaps this puts me in an elite echelon for this profitable hobby.

My counter-argument to Chris's objection: I've also gotten paid to teach others to draw stick figures. In fact, when I taught a class on cartoon art as a narrative device at the University of Wisconsin-Eau Claire, I taught two dozen students, most of whom had not drawn before, how to draw excellent, expressive stick figures. And the lesson barely took an hour!

"Fine," Chris said when I told him that. "But if you can teach people, can you teach ChatGPT?"

And that's when I felt truly embarrassed. Because in my last post, I didn't practice what Chris, Levi, and I preach at Elevate AI Coaching. I didn't use EAIC's full range of unique thought-partnership methods. And as a proud artist, I didn't have a proper prompt-versation with ChatGPT. I didn't enter the thread with an open mind and a desire to push the boundaries of what LLMs can do.

Contrite, I went back to ChatGPT and treated it like a thought partner, subordinate to me, but capable of learning new things... albeit in a synthetic way that doesn't involve true thought.

A person in futuristic armor with a full-face helmet, identified as The Mandalorian from the Star Wars show of the same title, stands against a bright sky. Text at the top states, "This is the Elevate AI Coaching Way," suggesting a strong, determined approach.

Starting the Experiment off Properly, with Priming

In one of my earlier art & culture focused articles on prompting (the one about Walt Kelly's Pogo comic strip), I talked about the importance of priming an LLM on a topic. What is priming? It's warming up by having a general discussion on a topic before you dive in to the main prompt. This helps the LLM "think" more clearly about the topic. It also helps you clarify your own thoughts as you move forward. Priming works best when you provide the LLM with context. Context takes two forms: detailed messaging about your own perspectives, and external context: uploaded documents and images, links to websites, and so forth.

I primed the LLM first by describing my perspective. I told it a little about my own "career" as a stick figure artist, linked to my previous blog post on this topic, and described Chris Lele's pushback.

The LLM offered to see if Chris was right, and try to learn how to draw stick figures imbued with human personality.

That's when I gave it a far more in-depth priming: a full art lesson. I started by giving ChatGPT the same thing I gave my students at the University of Wisconsin-Eau Claire: a model sheet of stick figure face expressions (seen below):

15 simply drawn circular faces with different expressions, numbered in order from left to right, top to bottom. 1: happy; 2: sad; 3: Angry: 4: Happy, excited; 5: sad, afraid; 6: angry/suspcious; 7: happy, gentle, warm; 8: surprised/scared; 9: nervous/sheepish; 10: skeptical/unimpressed; 11: relaxed/smug; 12: pained/upset; 13: laughing/amused; 14: thougtful/unsure; 15: playful/daydreaming

This is where my teaching method diverged from what I did with my human students. With the undergrads I served back in 2015, I merely told them to copy the face expressions into their own work and iterate in their own style.

But, as I painfully learned in my experiment in the previous blog post, AI does not have its own style. And it doesn't truly iterate-- it merely remixes things it's already been given. To get ChatGPT to remix well, more priming was called for.

To get ChatGPT "thinking" even more deeply, I asked it to identify, in 10 words or less, the emotions on each face in the model sheet.

It did a pretty good job! The LLM did initially describe face expression 9 as "Warm, friendly, affectionate" but when I asked it to reassess, it correctly re-identified the 9th face on the model sheet as "Embarrassed, awkward, nervous laughter, uncomfortable." For a task as complex and fundamentally human as drawing stick figures, I would still need to do far more priming, though. Next, I asked ChatGPT to analyze several of the earliest installments of my daily webcomic, Stick Figure Apocalypse. I had to go with the earliest episodes because my newer stuff is no longer drawn in stick figure style! To save you the trouble of clicking my webcomic and reading from the beginning (unless you want to), here are a few of the comics I dropped into the thread:

Two panel comic strip. Headline: Stick Figure Apocalypse, Altoona Beach Part 1: Therese and Matthew bring the thunder. First panel: Therese and Matthew are swimming in water and there are dark clouds above, with two lighning bolts. Therese says: Those clouds look ominous. We should probably get back to shore before the lightning comes. Matthew points to the lightning bolts and says "There is is." Second panel: Matthew and Therese frantically swimming away from the lightning. Both characters are stick figures. Threse is wearing a one piece women's bathing suit. Matthew is shirtless.

One panel comic strip. Header: Stick Figure Apocalyse: Ridin' with Caleb. Image: Matthew and Caleb standing in front of a truck with an open trunk. Caleb is holding up a piece of paper. Dialogue: Matthew: So I'm thinking, if the cops stop us, we can--; Caleb: Essential worker travel permit. Covers me, and anyone I'm with. Matthew: All righty, then.

Cartoon titled "Stick Figure Apocalypse" showing a scene outside Olson's Ice Cream. A stick figure with curly hair on a bike says, "For some reason, Olson's is open!" Another stick figure sitting on a bench replies, "Curbside pickup...?" The cyclist responds, "No! It's packed." The shop window displays several stick figures inside with ice cream cones. Text at the bottom reads "Downtown with Gabbey."

This is the summary ChatGPT gave me, in table form:

ELEMENT	YOUR STYLE
Drawing	Loose, expressive, intentionally imperfect lines
Characters	Distinctive hair/faces; emotion-led poses
Text	Handwritten, informal, bolded emphasis for comic timing
Emotion	Exaggerated, layered, visually reinforced through sweat drops, eyebrows, etc.
Humor	Observational, surreal, pandemic-era grounded, rich in voice
World	Recurrent cast with inside jokes and personality arcs

ChatGPT's analysis rang true to me, but I still had cause to be suspicious-- its analysis was laden with the effusive praise that OpenAI swears it's been teaching ChatGPT to stop doing. As I've mentioned before, beware the aggreeableness of LLMs.

So I had ChatGPT compare my work, side-by-side, with the work of the only truly famous stick figure cartoonist out there-- Matt Feazell. This helped ChatGPT understand both what I am doing and what I'm not doing, stylistically, and helped it break free of low-substance praise of my work.

I then had it re-describe my work without a direct Feazell comparison. The new description was better, but also a lot longer, and the tone of near-groveling praise was even more shrill, almost sickening. I'll spare you any quotes from that, and instead show you the visual style guide ChatGPT distilled from its description of my work:

A hand-drawn infographic on lined notebook paper titled “Expressive Stik Guide.” It showcases stick figure drawing traits. There are five labeled sections:

Linework: A stick figure with uneven arms and legs labeled “Wobbly, organic lines.”

Facial Features: A close-up head with wide oval eyes and dynamic eyebrows, labeled “Wide oval eyes, Dynamic eyebrows.”

Clothing: A smiling stick figure girl with yellow hair wearing a purple skirt and labeled “Distinctive skirt.”

Body Language: A panicked stick figure with arms flailing and bent legs, labeled “Exaggerated flailing arms” and “Hunched posture.”

Comic Panel: A colored panel of the same girl sitting hunched on a bench, yelling “I can’t believe this!” Her arms and eyebrows are expressive. She appears frustrated.

Sloppy and a little weird? Sure. But to my thinking, this was a good sign that ChatGPT "got it," at least as much as ChatGPT could. ChatGPT was now ready to draw in my style! Or at least, something adjacent to it.

If this priming process sounded a bit exhausting to you... well, I'll give you that. But some up front time investment in priming can pay off in the long run. And in this case, it did pay off-- just not in the way Chris Lele or I initially thought it might.

ChatGPT's Art Abilities After Priming

I gave ChatGPT two captions, both of which I had hoped could serve as attention getting text for LinkedIn posts, with ChatGPT stick figure illustrations.

The two captions were:

Can I teach ChatGPT to draw with my own unique visual voice?

AND

Don't let AI FOMO cloud your judgement.

The initial results nearly made me facepalm:

An AI-generated cartoon made to resemble a hand-drawn style shows a cheerful stick figure girl with yellow hair, a purple skirt, and a white shirt. She smiles with one hand on her chin and the other pointing upward. Large handwritten-style text to the left reads: “Can I teach ChatGPT to draw in my unique visual voice?” — though the top of the first line is slightly cropped off.

An AI-generated cartoon illustration in a hand-drawn crayon style. A worried stick figure girl with yellow hair and a purple skirt looks anxious, touching her face with one hand and gesturing with the other. A gray scribbled thought cloud appears above her head. Large red handwritten text to the right reads: “Don’t let AI FOMO cloud your judgement.”

STIFF. BOILERPLATE AI! BAD CHATGPT!

Sorry for my emotional overreaction. Caught up in my frustration, I forgot a very important fact: ChatGPT does not truly think and learn. It is merely a predictive text model on steroids. There is no point in getting frustrated with it. Imagine you are truly "teaching" and it is truly "learning" if that helps. But as needed, give yourself a reality check-- all you are doing is nudging predictive outputs.

Nudge, I did. I gave ChatGPT my own versions of the images it had created. Here are my very human drawings:

A hand-drawn crayon-style illustration on white paper. On the left, a stick figure girl with yellow hair, a purple skirt, and pink shoes stands smiling with one hand on her chin and the other outstretched. To the right, in large purple text, it reads: “Can I teach ChatGPT to draw in my unique visual voice?”

A hand-drawn crayon-style cartoon on white paper. A stick figure girl with yellow hair, a red skirt, and blue shoes appears anxious. She’s sweating, hunched forward, and glancing sideways. A black scribbled thought bubble above her head says: “ACT NOW.” To the left, bold red text reads: “Don’t let AI FOMO cloud your judgement.”

Time for some more priming. I told ChatGPT to "think" carefully about its version and mine, and the differences. It suddenly became its own greatest critic! Here are a few of the things ChatGPT said about its own art: generic... controlled... emotionally smoothed... emotionally distant... lacking spontaneity... robotic... staged rather than reactive... lacks visual "voice" It then made a list of things I can do that it can't:

✅ What You Achieve That I Don't (Yet)

Visible human effort → You convey a character who isn't just feeling something—she was drawn by someone feeling it.
Imperfection with purpose → You leverage wobble, asymmetry, and rough edges to tell a story. These imperfections feel alive.
Expressive exaggeration → Your characters express too much, and that’s perfect for humor and relatability.
Voice-match between drawing and text → Your captions and drawings are in perfect sync tonally. You write like you draw.

Finally, ChatGPT asked if I would like it to "bridge the gap," using its own self critique to create something closer to what I drew. Here are comparisons, side-by-side:

A side-by-side comparison of two cartoon illustrations with the caption “Don’t let AI FOMO cloud your judgement.” On the left is a hand-drawn crayon-style image labeled “My original drawing,” featuring a worried stick figure girl with yellow hair, a red skirt, and blue shoes, surrounded by sweat marks and a scribbled thought cloud that says “ACT NOW.” On the right is a cleaner, AI-generated imitation labeled “AI imitation,” with a similar girl in a slightly different pose and style, also with a thought cloud that says “ACT NOW.”

A side-by-side comparison of two cartoon illustrations with the caption “Can I teach ChatGPT to draw in my unique visual voice?” On the left is a hand-drawn crayon-style image labeled “My original drawing,” showing a smiling stick figure girl with yellow hair, a purple skirt, and pink shoes, with one hand raised and the other touching her chin. On the right is an AI-generated imitation labeled “AI imitation,” featuring a similar pose and expression but with a smoother, digitally styled appearance and blush on the cheeks.

If you were to take the images on the right and take them completely out of context, they actually look pretty good? They don't have the personality of the originals, but they have significantly more personality than the average AI "art."

In context, however, this isn't helpful at all. I had to spend 13 minutes creating the first two images, and it took me about 20 more minutes to get ChatGPT to figure out how to do good-but-weaker imitations of them.

ChatGPT Has Learned One Human Skill: The Ability to Use the Word "Yet" Ominously

Let's revisit "What you achieve that I don't (YET)." That yet sounds a lot more ominous when you look at my drawings above and realize how easily others could use ChatGPT to make passable plagiarisms of cartooning from real humans. This is not just paranoid conjecture. The Daily Cartoonist notes that someone with the handle "ToonAmerica" is using AI exactly the way I did to create weaker imitations of my own art... but they're imitating OTHERS' art-- the proprietary drawings of established political cartoonists! Here's one example (which I originally saw in political cartoonist John Kovalic's recent Facebook post):

A compilation of three side-by-side comparisons between original political cartoons and AI-generated imitations labeled "AI FAKE" or “AI imitation.”

Top row (Clay Bennett): On the left, a woman in a patriotic sweater clutches an empty wallet as dollar bills fly away under the heading “LIBERATION DAY.” On the right, a similar but distorted AI version copies the layout and message with exaggerated facial features and altered text styling.

Middle row (Tim Campbell): The original shows a FedEx-style truck labeled “FedWREx” with Trump driving and the caption “When it absolutely positively needs to be screwed-up in the first 100 days.” The AI imitation recreates the truck and driver with less precise linework and slightly altered text and proportions.

Bottom row (Rob Rogers): The original cartoon depicts Elon Musk with a chainsaw labeled “Musk,” walking away from dismembered social media logos and government agencies, saying, “Now I need to go save Tesla!” The AI version closely mimics the scene but simplifies the style and changes text slightly to: “My work is done here!”

Each original cartoon is clearly labeled with the artist’s name, while each imitation includes visible distortions and the watermark “ToonAmerica.”

Gloom and doom aside, this whole adventure paid off in a surprising way. AI is still NOT the most effective way to create unique eye-catching art, but....

ChatGPT Has Learned Another Nice Human Skill: Art Critique

In spite of the failures I've outlined, I can still say that my exchange with ChatGPT led to some really nice art. How did that happen? Astute readers may have already noticed a tiny detail. If you haven't spotted it yet:

Illustration comparing two similar images. On the left, a worried stick figure woman with a storm cloud over her head and the text "Don't let AI FOMO cloud your judgement." On the right, a similar drawing with a larger, darker cloud reading "ACT NOW" and an arrow pointing to it, emphasizing urgency. The left image is labeled "AI original" and the right "My remake."

The cloud wasn't my idea originally. My version is still definitely better. But it seems that AI, when adequately primed, has some pretty good ideas about art, even if it executes those ideas poorly. I had a hunch: ChatGPT may not have learned how to draw truly well, but it seemed to have learned how to analyze and critique art!

Sure enough, at the end of this long thread, I asked ChatGPT to revisit this drawing and tell me how it could be better:

A comic titled "Stick Figure Apocalypse" featuring a scene outside Olson's Ice Cream. A person on a bicycle says, "For some reason, Olson's is open!" Another person asks, "Curbside pickup?" The cyclist replies, "No! It's packed." Through the window, a crowded ice cream shop with stick figure customers is visible. The caption at the bottom reads "Downtown with Gabbey."

Here's the excellent advice I got:

🛠️ Opportunities for Improvement:

In one panel, Gabbey’s legs are completely straight, which feels more rigid than her emotional tone suggests—adding a subtle bend would show tension.
Her arms sometimes lack gesture—bringing hands closer to the face or chest would amplify her emotional stakes.

Not perfect (in reality one leg is straight, and one arm is stiff), but still rings true and is actionable. And this advice definitely came from good in-thread priming. I know this, because I started a new thread that lacked memory of this main experimental thread. The advice I got in the new thread for the exact same cartoon was much more muddled:

⚠️ Weaknesses & Areas to Improve

1. Inconsistent Line Quality / Density: Some elements (like the bike spokes and the brick wall) are drawn with much more line detail than others. This variation can pull attention away from the characters and story. Consider simplifying or stylizing these background elements to match the expressive minimalism of your figures.

2. Speech Bubble Layout & Flow: The dialogue could be slightly confusing at first glance due to the placement of the speech bubbles. “Curbside pickup...?” is visually in between the two characters and may be mistaken as coming from the girl. Consider flipping the order or adjusting tail positions for clarity.

3. Facial Style Inconsistency: Gabbey has more expressive eyes and eyebrows, while the biker has more realistic hair and freckles, creating a subtle visual disconnect. Simplifying the biker’s features slightly (maybe removing pupils or softening the freckles) could unify the look without losing personality.

4. Background Clutter Inside Olson’s: While the "packed" interior is key to the punchline, the crowd scene is busy and could benefit from slightly clearer staging — perhaps grouping the figures in overlapping clumps or giving a bit more negative space around the window edges to guide the eye.

You can read that carefully, or just trust me as an elite stick figure artist-- the unprimed advice makes no sense!

The Takeaway: How to Partner with AI to Create Eye-Catching, Easy-to-Produce Art

The two stick figure images I drew, scanned, typed, and and colored myself each took less than 7 minutes to create. If I had gotten some advice from ChatGPT afterward and used it for a redraw, maybe that would add 10 minutes to each drawing. 16ish minutes for great art that is AI-aided but doesn't look like the other AI slop out there? That's a great deal, especially since it would likely take you more than 16 minutes to prompt an almost-decent piece of LLM slop!

So, to partner effectively with AI for art, make your own art. Make it fast, make it simple. Stick with stick figures. Then get a critique. And if you are not (yet) the kind of elite stick figure professional that I am, prime ChatGPT with your thoughts and external context on stick figure art, and use it as your personal stick figure tutor.

As a former teacher, I generally disagree with the statement "Those who can't do, teach." But in the case of ChatGPT, it's true. Lousy artist, but not a bad artistic thought partner and stick figure art tutor.