How to Make Print-on-Demand Designs With AI (My Workflow)

You can make a print-ready print-on-demand design with AI in three moves: write a tight prompt, generate the art with a tool that handles your style well, then export it as a transparent, high-resolution PNG. The part most tutorials skip is everything after “generate a cool image,” which is exactly the part that decides whether your file prints clean and actually sells. I’ve uploaded thousands of designs across Amazon Merch, Etsy, and my own Shopify store, so here’s the workflow I actually use, including the steps that cost me the most time when I got them wrong.

Pick the engine that fits the design

The tools are not interchangeable. I switch based on what the design needs.

Ideogram is my default for anything with text: quote shirts, “nurse life” mugs, bold typographic designs. It spells words correctly, which Midjourney still fumbles. It’s cheap, roughly a cent or two per image on a paid plan.
Midjourney ($10/mo to start) makes the nicest illustrations, but it’s bad at text and it won’t hand you a transparent file. Good for art, annoying for production.
Nano Banana Pro (Google’s Gemini image model) gives clean, high-detail output and follows instructions well, which matters when you want a specific layout. It costs more per image than Ideogram.

My rule: text-driven design, use Ideogram. Illustrated sticker or detailed art, use Midjourney or Nano Banana. Don’t marry one tool.

Write a prompt that prints, not just one that looks good on screen

A design can look great at 1000px on your monitor and turn to mush on an 11-inch chest print. Prompt for print from the start:

Name the style plainly: “flat vector sticker, bold black outline, limited color palette.”
Ask for a transparent or plain solid background so you can cut it out cleanly.
Avoid fine gradients and thin lines. They band and disappear in DTG printing.
Keep the subject centered with breathing room, not bleeding off the edges.

A prompt I’d actually use: “A sleepy corgi wearing round sunglasses, flat vector sticker style, bold outline, 4 to 5 flat colors, centered, plain background, no text.” Specific beats clever. Vague prompts give you vague, unsellable art.

The boring step that makes or breaks the file: transparent and print-ready

This is where beginners lose hours. For most POD products you want a transparent PNG so the design sits on the shirt color instead of a white box. You also need enough resolution: aim for a long edge around 4000px so you’re comfortably above 300 DPI on a standard print. I cover the sizing and the no-Photoshop way to do it in how to get a transparent background for t-shirt designs.

If you generated on Midjourney, you’ll be removing the background yourself or paying a tool to do it. That background-removal dance, three tools to make one finished file, is the reason I eventually built LzyPOD to do the design and the transparent cutout in one step. The workflow here works with whatever tools you like, though.

Make mockups for ads and listings (without a photographer)

A flat PNG doesn’t sell. A shirt on a person does. Once you’ve made the product at your supplier, take that product photo and turn it into lifestyle mockups: someone wearing the tee, the mug on a desk, the tote on a shoulder. I used to do this by hand in ChatGPT and Gemini, uploading a photo and re-prompting every single time, which got old fast. However you make them, your Etsy listing and your Facebook ads both convert better with a real-looking scene than with a flat file on white.

Test before you fall in love with a design

The biggest money-saver in POD isn’t a tool, it’s killing losers fast. I don’t agonize over which design is best. I make a batch, list them, and let the market vote. For paid traffic, $5 to $10 a day on a Facebook ad for a few days tells you plenty. For Etsy, watch which listings get favorites and clicks in the first week. Most designs do nothing. That’s normal. The job is to find the few that move and pour more into those.

This is the whole reason I optimize for speed over perfection. If a design takes me 20 minutes, I test fewer ideas. If it takes 20 seconds, I test 50 a week and let data pick the winners.

My current stack, honestly

Right now I generate most designs with an AI engine, get the transparent file automatically, make lifestyle mockups from my product photos, then push the winners harder with ads. I built LzyPOD to collapse the design-plus-cutout-plus-mockup steps into one place because doing it across three apps was the slow part. You don’t need my tool to follow this workflow. You do need to stop hand-editing files and start testing more ideas.

If the part that’s slowing you down is the transparent-file and mockup busywork, that’s the exact thing LzyPOD automates. You can try it free and see if it saves you the same weekend it saved me.

Next, read can you legally sell AI-generated designs on Etsy before you list, so you don’t build a shop on shaky ground.