Can AI separate a simulated process job well enough to print?
I wanted a real answer to that, so I tested it. Images, the UK trade magazine for printwear professionals, just republished that article in its November 2026 issue. If you haven’t read it yet, this is a good time.
Why Did I Run This Test?
Shop owners keep calling me about AI-generated customer art. Your social feeds are probably full of complaints about “AI slop” files too.
Here’s my view. Customer art has always been a mess. We’ve printed from napkin sketches, blurry JPGs pulled off a website, and logos screenshotted from a text message. AI files are often better than that. They still need work, though, and the same tools that created them can fix them.
That got me wondering how far the tools could go. Fixing a low-res file is one thing. Simulated process separations are skilled work, and they’re the job art departments struggle to staff.
What Are the 3 Problems With Customer AI Files?
AI files show up with resolution that’s too low for production. They have errors like misspelled words, unwanted backgrounds, or strange anatomy. And they carry millions of colors when the customer wants a 3-color print.
The article shows how to prompt AI to fix each one, from upsizing to reducing the art to specific Pantone colors. That part applies whether you screen print, run DTF, or digitize for embroidery.
How Did the Sim Process Test Work?
I built a skull design with “SKULLFEST” lettering using Midjourney and ChatGPT. Then I wrote a Markdown instruction file with separation rules and gave the same file, art, and prompts to ChatGPT and Claude. Both ran on the lowest-tier paid plans. I refined the instructions after every round.
The target was a full-front print, 15" high by 12" wide, at 300 ppi, separated in channels, with the black shirt as a channel.
What Were the Results?
Claude consistently beat ChatGPT. Its first unrefined attempt was better than some seps I’ve seen from beginner separators.
The breakthrough came after I tightened the instructions. The AI stopped treating colors as adjacent tones that had to add up to 100% coverage, and it started thinking in layers and wet-on-wet printing. It also used the black shirt to build darker shades without adding screens.
The final output was about 8 channels, from underbase to highlight white. The separation took a little over 10 minutes. Total time from blank idea to finished separation was roughly 30 to 40 minutes.
Does This Replace Your Separator?
I’m not claiming that. Your designer still makes the creative calls. What AI does is speed up the repetitive part so your skilled people can spend their time where it counts.
You need to know what your art department costs you right now. Time each job and price out the hours. Then compare that to what you just read. Your competitors may already be testing this.