These days, one of my most frequent calls revolves around “What are we going to do about AI art?” I’ve spoken, emailed, or DM’d shop owners about this topic easily 20 times this week alone, as they want answers.

Online, on social media, and in Facebook industry groups, there is a constant barrage of people posting that they are “tired of working with AI Slop,” “I’m not accepting AI art,” or any of a hundred other statements about the problem.

Let’s Back Up a Sec

Just so you know, this article has a few solutions that will melt your brain, but before we get into that, I want to just throw something out there. If we rewind only a few years ago, before AI, you had a constant stream of customers trying to provide you with art ideas for their orders. This was a scribble on the back of a bar napkin. It was a 15k .jpg they downloaded from the internet. Maybe even something designed in Word.  

We’ve always had problems with customer art. Now there is a new villain to complain about. And this one actually sends us art files that are multiple times better than before, but you still might have to do something with them.

Didn’t You Hear the News?

Because our industry has been constantly complaining about getting these ChatGPT art masterpieces, if you missed the news, check this out:

  • Claude started working and connecting to Adobe products. Anthropic released 9 creative connectors on April 28, 2026, including Adobe Creative Cloud.
  • ChatGPT expanded with Adobe, with a ChatGPT plugin launched on August 6, 2026, but it consolidated 3 ChatGPT integrations launched in 2025 for Photoshop, Express, and Acrobat Pro.
  • Gemini added Adobe (Photoshop, Lightroom, Express, and Firefly) in September 2026, and it is still in early release.

There are a few issues I consistently hear about with these “AI Slop” files.  

  1. The most common complaint is that files created with AI tools are too low-resolution to use properly.
  2. Next, there could be something wrong with the file. A word is misspelled, the background needs to be removed, or there is something weird in the art, like 8 fingers on one hand.
  3. Lastly, when we are talking about common complaints for AI file usage, it is that there are too many colors in the files provided. The client sends an AI file with millions of colors, but wants it screen printed in 3 colors.

The point I’m trying to make here is that these issues are not impossible to solve. You can do it right now using the same tools that the art was created with.

All you have to do is connect these tools to your AI platform (you can just ask it how and your tool will walk you through what to do) and get started.

The AI Slop Art Fix

Let’s take the low-resolution file problem to start. Try this right now. Open up your AI tool of choice. Upload the file your customer sent you. Use this prompt:

“Here’s an image file. Please make this file 300 ppi at [size choice] wide and save as a Photoshop file.”  In about 30 seconds, your file will be at the right dimensions and resolution to use.

Next, let’s say there is something wrong with the file. (A word is misspelled, the background needs to be removed, or there is something weird in the art, like 8 fingers on one hand.) Don’t forget to fix those while you are at it.

“Here’s an image file. Spell “Volcano” correctly, and use the same font. Change the hand of the guy with the shovel to have 5 fingers instead of 8. Delete the orange background entirely. Make this file 300 ppi at [size choice] wide and save as a Photoshop file.” 

Lastly, on the common problem list, the file has too many colors. That’s just as easy.  

“Here’s an image file. Spell “Volcano” correctly, and use the same font. Change the hand of the guy with the shovel to have 5 fingers instead of 8. Delete the orange background entirely. Reduce the number of colors to 3. Use PMS 289 Navy Blue, PMS 116 Yellow, and PMS 186 Red. Make this file 300 ppi at [size choice] wide and save as a Photoshop file.” 

Bonus. Don’t like using Photoshop and need things in vector? No problemo.

“Here’s an image file. Spell “Volcano” correctly, and use the same font. Change the hand of the guy with the shovel to have 5 fingers instead of 8. Delete the orange background entirely. Reduce the number of colors to 3. Use PMS 289 Navy Blue, PMS 116 Yellow, and PMS 186 Red. Use Illustrator to live trace to vectorize the file and save the file as an Illustrator .eps file.”

This Is Where AI Is Going to Blow Your Mind

So, I’ve been playing around with this for a bit. I run AI experiments all the time to broaden my knowledge and find out what works. Here’s where it started. I asked myself:

“If AI can do all of those art tasks, what about handling something more complex like simulated process separations for screen printing?”

Yep, it can do that too. But it’s complicated.

Simulated Process Screen Printing Separations Using AI

Frankly, I could bore you to death with a report on all of the files I’ve run through AI and separated. I kept banging away at this and tried to understand what it needed to do a good job at the task. So, I started putting in rules and guidelines and created a Markdown file to use to guide the AI platforms. In my testing, I compared ChatGPT and Claude using the same images, Markdown files, and text prompts.

Right now, here is what I can report. Claude is the hands-down winner with this.

Go ahead. Try it. Take the last sim process job you worked on, upload it to Claude, and use this prompt:

“Separate this file for screen printing using Pantone colors. The final file should be a Photoshop file. It will be printed on a [color] shirt.”

That’s how I started. The result? Ok, but not perfect. Honestly, what it provided was better than some seps I’ve seen printed on press by beginner separators. I’ve been doing this since the late 1990s, so I’ve definitely got my 10,000 hours in on sim process seps. And I’ve been teaching this to my own staff, clients, and others in classes or workshops ever since.

I used a really difficult image to separate so I could understand the choices AI was making. Here’s my test file. By the way, the skull and the typeface were both designed in Midjourney. I assembled both in ChatGPT by uploading the image of the skull and the type alphabet, and asked ChatGPT to spell out “Skullfest” and emulate the skull image with a chrome look with reflected colors.  

This is the file I used in my major testing, and I told the AI tools to use this image, sep for a full front image at 15” h x 12” w, 300 ppi, separate in channels with a black shirt as a channel. I let the AI tools choose the Pantone colors they wanted to use for most of the trials.

Between the two platforms, Claude consistently provided better results. At the end of this article is the Simulated Process Separation Markdown file, which you can use to test this yourself. After each trial sep, I modified this Markdown file with tweaks on what I thought the AI missed, got wrong, avoided, or didn’t understand. 

Giving AI tools instructions about the outcome you want is essential to achieving better, more consistent results. Is the Markdown file I’m giving you perfect? No. I’ll be working on this as I keep experimenting or running the tools for projects.

Here’s the fact that I think you might like to hear. I got superior, professional-quality simulated process separation results that I could use with minimal extra tweaking in about 10 minutes. And for those of you curious, I’m on the lowest-level paid plan for both ChatGPT and Claude. 

Is it good enough to replace a very talented professional separator, or some software you are currently using? That’s a subjective question. Try it and you be the judge.

The CMYK Trap

Below are some detail shots of each of the 8 channels. Here’s what I want you to look for, as this is what thrills me. Before I tweaked the instructions, the AI tried to print only adjacent color tones, meaning color next to color, like in CMYK printing, where the tones had to add up to 100% coverage. Now, with the help of the Markdown file, it understands how simulated process seps use layering to achieve richer, fuller color blending by printing color on top of color, and that multiple color channels can have tone in the same areas. That was a breakthrough.

Also, if you handled color separations, you know you can extend the color range of your work by using the garment color (in this case, black) to modulate tones and create darker colors, shadows, and a fuller color spectrum without adding more screens.  

Below, I want you to focus on how Claude solved the problem of using the shirt color without tone in the underbase and of printing colors directly on top of the shirt color to create dimension. Compare the area between the skull's eye sockets on the underbase and on the other color plates, and see how it is dropping tone and building up colors to print wet-on-wet.

Here’s a screenshot of the sep file.

Here’s the Underbase White detail. The dark tone prints white, so keep note and compare where the shirt color will be used to modulate tonal ranges for darker colors on the print.

Here’s the PMS 803 Neon Yellow plate. Note how this is present under some of the other plates that will print afterward. Yes, Claude chose this color.

Here’s the PMS 2995 Light Blue plate. BTW, this Pantone color has always been a color I would choose for a rich sky blue. It made me happy to see that Claude chose it.

Here’s the PMS 806 Neon Pink plate. This color is built to be under many of the next plates to extend the blending and tonal range. Compare.

Here’s the PMS 021 Orange plate. Your basic orange. Note how it blends into the pink and yellow.

Here’s the Pantone Purple plate. I’ve always loved this color. It’s a great choice between the PMS 2995 and the PMS 2736 that prints after it.

Here’s the PMS 2736 Dark Purple plate. The darkest hue printed. Note how in some areas it is printing over the Underbase White, but also the black shirt without the underbase. Imagine how that will look and how it will provide more depth to the final print.

Here’s the Highlight White plate. After all the colors, and a flash to gel what has been printed, this goes down last to pop the white highlights and provide the upper end of the tonal range.

Here’s the detail shot of all of the channels turned on. One last look.

My Takeaway on AI Art Department Workflow

Believe it or not, I’m still a designer who makes choices, even when using AI tools.  

Yes, I used Midjourney to create the skull. But that was an iterative and curated process based on the prompt instructions I gave it. That took about 5 to 10 minutes to work through.

I used the same typeface technique for “SKULLFEST.” The alphabet I created was built on instructions I gave Midjourney. This was another 5 minutes to create.

Then, I assembled the downloaded skull image and the typeface alphabet sampler into ChatGPT and gave it instructions to lay out the design and create the word “SKULLFEST” using the Midjourney-created typeface. This took about 10 to 15 minutes because ChatGPT made a design error, and I had to iterate a few times to correct it.

Now, with a Claude process for separations, I can take an image I created and create the screen separations in a little over 10 minutes. From a time standpoint, this is exciting, as someone in your art department can go from idea to art assets to art approval to separations for a complex project like this in only about 30-40 minutes.

This isn’t about pushing creatives out of jobs. Instead, this is about giving them more creative time to do valuable work that makes your shop money by spending less time on repetitive tasks. Do a time or cost study in your art department to understand your current situation. Automating some of these steps is valuable.

If you aren’t playing with this, I can tell you that your competition surely is. As with learning to drive a car, the only way to fully understand it is to get behind the wheel and learn by doing.

Do yourself a favor. Grab the Markdown file below, pick a job or two you have already separated, and run a taste test. Double-check the files. Get what you want? Burn some screens. Run some test prints. See how this works. 

Lastly, I’m looking for a few shops that want to explore automating their art department workflows. I’m looking for explorers who want to go on a journey. I’m 100% going to be building out content and education around this. Want to be a part of that?  

Shoot me an email (marshall@marshallatkinson.com) or book a call.