Anyone can generate a beautiful image in thirty seconds. That is exactly why generating beautiful images is not a business. Print on demand rewards a different skill set: finding demand nobody has served, producing at the right resolution, and listing so buyers can find you. The art is roughly twenty percent of the work.
Step 1 — Find the niche before you make anything
The winning pattern is specific identity plus specific occasion. Not \”funny cat shirt.\” Instead: \”gift for a night-shift veterinary nurse who owns a three-legged cat.\” Narrow niches have less competition, higher intent, and buyers who forgive a higher price.
Research method that works: search marketplaces for a niche, sort by best selling, and look at review counts and dates rather than listing counts. Reviews mean money changed hands. Then check whether the top sellers are generic — if they are, there is room. Use Etsy and Amazon Merch for demand signal, and an assistant to cluster what you find into themes and phrase variations.
Step 2 — Design for the product, not the screen
The single most common failure is resolution. Print needs 300 DPI at physical size. A 15 inch wide shirt graphic is 4500 pixels across. Most generators do not output that natively, so upscaling is not optional — run finals through Magnific or Topaz before uploading.
- Text in designs: generate text separately in a vector tool. Model-generated lettering still fails at print resolution more often than it succeeds. Recraft handles vector output well.
- Transparent backgrounds: essential for apparel. Photoroom or a proper cut-out pass, not a rough auto-remove.
- Colour: what glows on screen prints muddy. Test one physical sample per product type before scaling a catalogue.
- Composition: design inside the printable area with margins. Mugs wrap, shirts have seams, posters get cropped.
Step 3 — Pick the fulfilment stack
Printful for quality control and branded packaging. Printify for margin and product breadth across competing providers. Gelato when international shipping cost is killing you. Amazon Merch when you want to fish where the buyers already are and accept the platform\’s rules.
Sell on a marketplace first for demand validation, then open a Shopify store once you know what sells. Building a store before you have a proven design is a very expensive way to learn about traffic.
Step 4 — Listings are the actual product
Two identical designs, different listings, produce wildly different revenue. Use a model to generate twenty title variants against the phrases real buyers search, then pick on judgement. Write bullet points that answer sizing, material and gift-occasion questions. Generate lifestyle mockups rather than flat images — a mug on a desk outsells a mug on white.
Volume without research is a spam strategy, and marketplaces have gotten very good at suppressing it. Twenty researched designs beat two hundred generated ones, and take less time.
Step 5 — The legal part, briefly
Do not generate anything resembling a trademarked character, brand, band or slogan — this is the fastest route to account suspension, and platforms suspend first and discuss later. Check each generator\’s commercial-use terms rather than assuming. Keep your source files and prompts; provenance records are becoming useful. And check whether your marketplace requires disclosure of AI-assisted work, because several now do.
A realistic first ninety days
- Weeks 1-2: research three niches, validate demand, produce ten designs in one niche.
- Weeks 3-4: list, order one sample, fix what the sample reveals.
- Weeks 5-8: double down on whatever sold at all, kill the rest, add twenty variations of the winners.
- Weeks 9-12: add a second product type on your best design, start collecting emails, and only then consider ads.
The full tool stack and prompt templates live on the image generation and print-on-demand page. If you would rather sell knowledge than objects, see writing and books, and for the promotion side, ads, marketing and social.
