Is AI Graphics Still Crap or Do I Just Suck?

Is AI graphics still crap or do I just suck? That’s the question I was asking myself after spending several hours trying to get some thumbnail graphics for this very blog. Guys, I’m not a novice to this AI stuff. I’ve been an early adopter (reluctantly or not). Granted, I’m not a super-uber nerd that lives and dies by what’s going on in the AI world, but I’m certainly more trained and more experienced than most. Especially in our local Ohio Valley market.

And I just couldn’t get a product that I was proud enough of to hit publish.

Let’s dissect this to find out if I am just bad at prompting or does AI graphics still have a bit of a way to go.

Am I Anti-AI Art?

If you hang out on my Twitter (X) feed, you will think I’m grouped right in there with the artists that are super anti-AI art. Well, it’s probably because I give most of their posts a like and often engage. Almost all artists have some valid points. Firstly, AI art used most of their graphics as inspiration (which is a nice way to say they stole it) and are now able to produce the same style, faster and without limited cost. That troubles me as a designer (and I’ve gone on this rant before so I will spare you).

But here’s the thing - I’m not 100% against AI art. Would I use it? Eh, probably not. But do I think some of the illustrators out there were getting a bit ridiculous with costs and turnaround times...probably. Not that they aren’t worth their cost, but technology evolved and things are a lot easier and faster to do.

In my opinion, Canva changed the game.

So saying a quick change on a graphic was going to be a week turnaround and cost me $200 per hour, eh, maybe I’ll just do it myself. That’s partially the reason we are here.

And I get it. The same thing happened in the web space. Wix, SquareSpace, and even WordPress made the $20k websites turn into $5k websites. Well, to build at least. The institutional knowledge and effectiveness of a professionally done website still has that premium.

And I think that’s where AI art gets it wrong and why designers still are necessary. AI art can craft up something quick - but it’s pretty generic. And the end user isn’t educated enough in the craft to know how to prompt to make it look different, better, or even (sometimes) good.

Making Some Trapper Keeper Images

One day I was scrolling through Facebook and ran into a graphic that caught my eye. Remember those old illustrations that were often found on 90s folders? They usually had some type of sport, done to the extreme, and usually a crazy detailed (and enlarged) animal? Yes, someone shared a post of some of these classics and it immediately sparked some nostalgia.

Well earlier in the week, I gave my friend some shit for posting an AI poster on Facebook. I then proceeded to make a bunch of equally crappy AI posters - themed around obnoxious items - and sent them via text for the next hour. My favorite was a Saved By the Bell-themed poster that had nothing to do with what he was promoting, but I feel like I made my point. Let’s do a little better.

So I was inspired to see if I could actually create something in AI that didn’t look like complete AI slop. And I decided to use the folder graphics as inspiration. I uploaded the examples and then started to craft my prompts to see how close I could create (rip off) the art style.

Maybe after an hour of killing time (while watching some TV), I managed to create something that was pretty close. Not as good as the originals, but at least they were passable. But it took some solid prompt work. I had to be pretty explicit with what I was asking. And if you had one little thing wrong, the whole darn picture would morph slightly just to fix the one issue.

My little fun project was done and I never looked back.

Ripping Off Myself

So I think philosophically I am opposed to stealing other people’s work. Besides the fact that I’m just not a fan of AI-created graphics, I know how these models were created so it’s hard for me to go all-in. But...what would I care if the art was ripping off my creations...for me? I’ve been perfectly fine letting AI use my writing to help craft posts in my voice (with modifications), so I wondered if I could get AI to take my recent thumbnails and recreate them.

Over the years, I’ve created hundreds of featured images and thumbnails, but more recently we’ve created 25-50 on the Strong Minded blog. They all have some similar features, so could we train the AI engines to take our examples and help create thumbnails for upcoming posts? If the quality was good enough, it could save me 10-15 minutes (or more).

So I started. I grabbed our style guide and all of the examples. I was going to diversify and try ChatGPT and Google Gemini. I worked on some decent prompts and started working.

Woof. They struggled.

A Few Lessons I Learned from AI Graphic Creation

To cut to the chase, we aren’t even close to having something I would release on our site. I don’t find our brand to be overly complex, but neither ChatGPT nor Gemini could quite understand what we were doing.

Issues with Font

The biggest issue is that ChatGPT could not get the font right. Turns out, they can’t take fonts from online and recreate them. This explained a lot. This is why all posters created have the same looking font. This is the biggest giveaway when I’m looking at graphics. Even when I pointed them to Google Fonts for Raleway (our font), they could not recreate it. Gemini did much better, but it was still slightly off.

Human Photos

I also uploaded several photos of the team. Maybe I need to upload more...but the results were pretty rough. You could tell that it wasn’t us. Both platforms would adjust our bodies in ways that seemed unnatural. I was just hoping to use the photo and crop out the background...but nope - they wanted to change far more.

Small Edits

I mentioned earlier - if you like everything but one thing, good luck. When you ask for the small change, something else will change. So unlike Canva or Photoshop, you can’t just slightly adjust a graphic or text. You have to prompt your way to fixing it, and that often creates a whole different version.

It Will Get Better (and Probably Has)

Honestly, we are way further along than I would have thought compared to a year or so ago. I dabbled with AI graphics back then and they weren’t nearly as sophisticated. You couldn’t use words at all.

But I realized in this example, I could create exactly what I wanted in less time in Canva or Photoshop than it would take me to prompt away with ChatGPT or Gemini (not counting the prep work with the style guide). But, I know things like Magic Layers in Canva are going to help fix some of the problems I complained about. And I’m also sure there are some AI tech gurus that have fixes for some of my problems.

But for the sake of this blog and those in the future, I will just roll up my sleeves and create for myself. In a year or so, I will try to steal from myself again.