Use a generator that can draw a tileable image, ask for a flat, evenly spread motif with no focal point, then test the result by tiling it three by three and repainting any visible joins. Export at the size and resolution your fabric printer asks for, and read both the generator's terms and the printer's rules before selling anything printed.
The step most owners skip is the test. Generators produce pictures that look like patterns, and many of them are not true repeats: the edges do not meet, a large motif is cut in half, or a faint gradient makes every tile visibly different. A pattern that looks fine on screen at full size can show stripes the moment a printer lays it across five metres of cloth. The work is mostly in checking, not in prompting.
What a print repeat is, in plain terms
A repeat is one rectangle of artwork, called a tile, designed so that copies placed edge to edge look like a continuous surface. Whatever leaves the right edge must re-enter on the left at the same height, and the same goes for top and bottom. The layout of the copies is the repeat type:
- Basic (block) repeat. Tiles sit in a plain grid. Simple, but a strong motif can show lines of identical shapes running across and down.
- Half-drop. Every second column drops by half the tile height. It breaks up visible lines and suits larger motifs.
- Half-brick. Every second row shifts sideways by half the tile width. A similar effect to half-drop, rotated.
- Mirror. Tiles are flipped alternately. Symmetrical and easy to join cleanly, but florals and text look odd mirrored.
Fabric print-on-demand services often let you pick the repeat type after upload, so one tile can give you more than one look. Spoonflower's help centre, for example, lists five options (basic, half-drop, half-brick, centre and mirror), chosen from the design's edit page, with centre available for fabric only. If you are planning a whole range, a tutorial on drafting tech packs with AI shows where the finished print file is referenced for the factory.
Choosing a generator: what to check before you pay
There are three kinds of tool a small brand might use, and the checks differ by kind.
- General image generators such as Midjourney, ChatGPT's image tool or Adobe Firefly. They draw anything, and some advertise a tiling setting or accept a request for a tileable image, but you cannot count on a clean join. Treat their output as raw artwork.
- Pattern-specific generators sold as textile design tools. They advertise tileable output and sometimes export repeat variants. Test the claim before you trust it.
- Design suites with pattern tools (vector editors with a pattern mode, or an image editor's offset command). These do not generate art but make any image into a verified repeat. Many small brands pair a generator with one of these.
Whichever you pick, answer these questions from the tool's own terms and pricing page, not from a review:
| Question | Where to look | Why it matters |
|---|---|---|
| Can I use the output commercially on my plan? | Terms of use, plan table | Some free tiers limit commercial use or make images public |
| Who owns the output? | Terms: ownership or licence clause | Decides whether you can license or sell prints exclusively |
| Is my prompt or image used to train models? | Privacy or data controls | Matters if you upload original artwork or brand references |
| What is the maximum export size? | Help centre, upscaling options | At 150 DPI, a 1,024-pixel tile prints only about 17 cm wide |
| Is there a tile or tileable-output setting? | Parameter list or settings | Saves repair time, but still needs testing |
| Does it have a stealth or private mode? | Plan table | Stops competitors browsing your unreleased prints |
Plans and allowances change several times a year, so I have not listed prices here. Open the pricing page of the tool you like on the day you buy, and note the date you checked in your own records.

The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowA main worked example: a botanical print for a linen scarf
Illustration only. Say a three-person accessories brand wants a pale botanical print for a linen-blend scarf. They have a mood board of six dried-flower photos taken from their own garden and a palette of four colours: sage, cream, rust and charcoal. Nobody on the team is an illustrator.
They start with a prompt that describes the pattern as a surface, not a picture:
Tileable surface pattern for fabric: loosely scattered dried wildflower stems,
hand-painted gouache look, flat colour, soft edges. Palette: sage green (#9CAF88),
cream (#F3EBDD), rust (#B5543A), charcoal (#333333). Cream background.
All elements similar scale, evenly spread, no focal point, no large empty areas,
no text, no border, nothing touching the edge of the image. Square tile.
Three details do the work. "Nothing touching the edge" avoids the hardest case, which is a stem that has to continue on the opposite side. "Evenly spread, no focal point" stops a big central flower that would show as a dot every tile. And the hex codes keep colours near their brand shades, though generators drift, so they re-check colours later in an editor.
The first result is pretty but has a big pale patch in the lower left and a cluster of rust flowers top right. Tiled, those would show up as a repeating blotch. They generate eight options, pick the most evenly spread, and move to the repair routine below.
The repeat check and repair routine
This is the part that turns a nice picture into a printable pattern. It takes about fifteen minutes the first time and five once it is habit. Use any editor with an offset function. In the free GIMP it is Layer, Transform, Offset, which has a one-click "Offset by x/2, y/2" button and a wrap-around edge setting that together do step 1 for you. Paid editors have an equivalent filter, and an online pattern-preview tool will do step 5.
- Offset by half. Shift the image half its width right and half its height down, wrapping around the edges. The old edges are now a cross in the middle of the picture, which is where joins show.
- Look for the cross. Any hard line, colour step or cut-off shape along that cross is a seam. Zoom to 100 per cent and scan.
- Repaint the cross. Use a clone stamp, healing brush or a copied patch of nearby background. Add a stem or two across the seam in the same style to hide it. Keep new elements to the same scale.
- Offset again. The repair should now sit near the old edges, and the new join should be clean. Repeat until the cross is invisible.
- Tile it three by three. Make a 3x3 grid of the tile in a new canvas. Zoom out to thumbnail size and look for stripes, diagonal lines or repeating blobs. Anything you can spot in a thumbnail will be obvious on a scarf.
- Try half-drop. Shift alternate columns by half the tile height and view again. Many patterns that stripe in a basic repeat look natural in half-drop.
- Check the colours. Sample the four palette colours and compare them to your brand values. Adjust levels or hue in the editor, not in a new prompt, because a re-prompt changes everything else too.
For the scarf, the team finds the offset cross crosses an empty patch, repaints it with two stems copied from elsewhere in the tile, and then notices in the 3x3 view that a rust cluster repeats on a diagonal. They erase two flowers from that cluster and the diagonal disappears. Total time: about twenty minutes, on top of ten minutes generating options.
More prompts for common print jobs
The same structure adapts to other briefs. Here are four more, each with the usual failure to look for:
- Geometric print for a tote bag. "Tileable geometric pattern of overlapping circles and quarter-arcs, two-colour (navy and off-white), flat vector look, even density, no gradients." Failure: arcs that stop at the tile edge instead of continuing. Fix with the offset routine, or redraw the arcs in a vector editor where joins are exact.
- Ditsy floral for children's wear. "Tiny scattered daisies and leaves, random rotation, even spacing, pastel palette, ditsy scale, no overlap with the edge." Failure: flowers cluster and leave gaps. Generate larger and downscale, since density looks more even at small scale.
- Abstract brush strokes for a cushion. "Large loose brush strokes in terracotta and cream, flat, same weight, continuous and wrapping at the edges." Failure: one dominant stroke. Ask for "no stroke longer than a third of the image".
- Stripe or check. Do not use a generator for precise stripes. Build them in a vector editor or spreadsheet grid. Generators wobble line widths, and a wobble in a stripe is a visible fault.
When a design must look hand-made, a useful trick is to scan a few of your own hand-drawn marks and add them on top of the generated tile. That adds individual character and strengthens your position on who made the artwork.
Printer file specs: read the printer, not the generator
The printer decides the file. Spoonflower's help centre, for instance, asks for JPG or PNG files under 40 MB, saved at your print dimensions and 150 DPI, in the sRGB colour mode it prints with. It converts every upload to 150 DPI and works out the printed size from the pixel dimensions divided by 150, so a file saved at 72 DPI prints at under half the size you set. Treat that as an example of what to look for, not a rule for every printer. A direct-to-fabric digital printer, a screen-print mill and a sublimation transfer service all have different needs.
Collect these answers from your printer in writing before you finish a tile:
- File type and colour mode (RGB or CMYK, any colour profile).
- Resolution, and whether the physical size is set in the file or by their upload page.
- Maximum file size and maximum pixel dimensions.
- Maximum repeat size, and whether they handle half-drop themselves.
- Number of colours allowed, if it is screen-printed.
- Whether they ask about AI use or content rights at upload.
A quick sum shows why resolution matters. A 150 DPI printer and a 24 cm square repeat means a tile of about 1,417 pixels square (24 divided by 2.54, times 150). If your generator exports 1,024 pixels a side, a 24 cm repeat needs upscaling or a larger export setting, so look up your own tool's export size, which varies by tool and setting. If instead you want a 12 cm repeat, 709 pixels suffice, and a plain export is enough. Check your own printer's figures against this working before you hit upload.
Upscalers add detail that was not in the original. Check an upscaled tile for odd textures at 100 per cent, and for broken joins, because an upscale can reintroduce an edge mismatch you had fixed.
Ownership of AI output and what to verify
I am not giving legal advice, and the rules differ from place to place, so this is a checklist of what to look at, not a ruling.
- The tool's terms. Find the clause about output. Some tools assign ownership of output to you; others give you a licence; some limit commercial use on a free plan. Save a copy of the terms page on the day you used the tool.
- Third-party content. If you uploaded a reference image, such as another brand's print or a stock photo, the output may be a derivative of it. Use only references you own or have licensed.
- Style and artist names. Do not prompt "in the style of" a living illustrator or a named brand for something you plan to sell. It invites complaints, and marketplaces may remove listings.
- Protection of the result. Whether purely AI-generated art can be protected by copyright is unsettled and varies between places. If a print is central to your brand, add your own substantial hand work, keep your working files, and ask a qualified adviser what registration or contract terms suit you.
- Marketplace and printer rules. Some sites require AI-made designs to be labelled: Etsy's creativity standards, for example, ask sellers to disclose AI use in the listing description. Print services can set their own content rules as well. Read the rule page for each place you list.
- Customer disclosure. Describe the product honestly. "Print created with AI assistance and finished by hand" is accurate and short.
Keep a simple provenance log for each print: tool and plan, date, prompt, selected output, edits made, and a link or copy of the terms at that time. It takes two minutes per design and answers almost any later question from a buyer, a marketplace or a printer.
A filled-in log entry for the scarf, so you can see the level of detail that is enough (illustrative):
Print: BOT-01 dried wildflowers, scarf and cushion
Date: [date of generation]
Tool and plan: [tool name], [plan name], terms page saved as PDF on the same day
Prompt: saved in file BOT-01_prompt.txt
Chosen output: option 5 of 8, 2048 x 2048 px
Edits: offset repair of cross at centre; removed two rust flowers (diagonal); rust adjusted +4 hue
Own elements added: three hand-drawn stems, scanned from the sketchbook
Printer check: 150 DPI, JPG, under the size limit, half-drop selected
Proof: swatch ordered, rust too orange on cotton, corrected in v2
Disclosure line: "Print created with AI assistance and finished by hand."
Six months later, if a marketplace asks how a design was made, this one block answers it in a minute.
A proof before you print the full run
Screens lie about fabric. Order a small test, such as a fat quarter or a half-metre swatch, before committing to a roll. When it arrives, check:
- Do colours match your palette in daylight, not only under shop lights?
- Do any lines or blobs show across the repeat on a metre of cloth?
- Is the scale right on the actual product? A motif that looked delicate on a screen may be oversized on a small scarf.
- Does the fabric type change the look? Light prints on dark base fabrics, or on textured linen, behave differently.
Say the scarf proof shows rust looking more orange than on screen. They adjust the file's rust by a few points, reorder a second swatch, and approve. Without the proof, they would have found out on a full run.
A repeatable routine for a collection of four prints
When you move from one print to a mini-collection, consistency matters more than any single tile. A routine that works for a small team:
- Write one brand brief: palette with codes, scale rules, mood, a list of forbidden things (text, faces, logos, named styles).
- Save it in a Project or custom instructions so every prompt starts the same way. The tutorial on AI marketing tactics for fashion brands uses the same idea for tone of voice.
- Generate 8 to 12 options for each print, shortlist two, and run the repeat check on those.
- Build coordinates: a large-scale print, a small-scale companion and a simple texture, all in the same palette, so pieces mix.
- Run the printer proof, record the provenance log, then photograph the swatches for the listing. A guide on making product photos with AI on a small budget helps with that last step.
Colourways are a better job for a chat assistant than for an image generator, because the request is about colour, not drawing. An illustrative exchange for the scarf collection:
My brand palette is sage #9CAF88, cream #F3EBDD, rust #B5543A, charcoal #333333.
Suggest three colourways for a botanical scarf print and a matching small-scale texture.
For each: background colour, two motif colours and one accent, using only my palette
or tints of it (state the tint percentage).
A typical reply proposes "Day" (cream ground, sage and charcoal motifs, rust accent), "Dusk" (charcoal ground, cream and sage motifs, rust accent) and "Clay" (a 20 per cent tint of rust as the ground, with cream and charcoal motifs). The first two can go straight to a recolour. The third needs a swatch before anyone relies on it, because rust already drifted on the first proof. Recolour the repaired tile in your editor for each colourway rather than generating new art, so the joins you fixed stay fixed and the motifs line up across the set.
If something fails the repeat test twice, switch approach rather than re-prompt: draw the motif in a vector editor and let AI suggest colourways instead. AI is better at variations on a design you control than at inventing a flawless repeat from nothing.
Six ways an AI print fails once it is on cloth
- Judging the tile on its own. A single tile never shows the stripe. Always view the grid.
- Re-prompting to fix a seam. Each new generation changes everything. Repair the good one.
- Exporting at screen size. The printer rejects it or prints it soft.
- Trusting the colours. Screens use light; fabric uses ink. Print a swatch.
- Skipping the terms. A free plan that forbids commercial use is a problem found after the full run is printed.
- Selling a design you cannot show you made. No log, no prompt, no working files. Keep them.
The honest summary: a generator can give a small brand a faster route to many print ideas, while the repeat check, the printer's specification and the ownership questions still decide whether any one of them can safely go on cloth.
Print repeat questions that come up with AI patterns
Why does my AI pattern show lines or blobs when tiled?
Most generators draw one picture that looks continuous at its edges only by chance. When the tile repeats, the left edge meets the right edge and any mismatch appears as a seam. Offset the image by half its width and height, repaint the join, and test again on a three-by-three grid.
What resolution should a repeat tile be?
That depends on the printer. One print-on-demand fabric service states that it prints at 150 DPI and wants JPG or PNG files under 40 MB, with the printed size set by pixel dimensions divided by 150. Your printer may differ, so read its own help page before you export.
Can I stop others copying my AI-made print?
Possibly not as strongly as with a hand-drawn design. Protection for work made mostly by a generator is unsettled and differs between places. Your own edits, hand-drawn elements and arrangement help your position, but ask a qualified adviser before you rely on exclusivity.
Do I need to tell customers a print was AI-assisted?
Some marketplaces require it: Etsy's creativity standards, for example, ask sellers to say so in the listing description. Some customers care too. Check the rules of each place you sell, and when in doubt, say so plainly in the product description. Honest labelling costs little and lowers the risk of a listing being taken down.
Further reads
- AI Tools for Ecommerce Photography: Backgrounds, Shadows, Consistent Sets — Clean product shots for the printed items.
- AI Videos for Clothing Brands: Reels From Product Photos — Turn swatch and product photos into short reels.
- AI Tool for Clothing Designers: Tech Packs and Mock-Ups Before Sampling — No single AI tool covers sketch to sample: which tool does each stage, what a tech pack draft needs from you, and what to leave to the pattern cutter.
- AI Design Software for Clothing: Mood Board to Sample Brief — Move from inspiration to an annotated sample brief, with a practical check on generated flats, colours and ownership.
- AI Marketing for a Clothing Brand: One Drop Planned Start to Finish — Plan a complete clothing drop from teaser to restock using ChatGPT and Canva with exact inventory numbers and asset schedules.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Spoonflower help centre (Uploading Images, 150 DPI, Repeat Options); GIMP documentation for the Offset command; Etsy creativity standards; each generator's own terms, to be checked by you before commercial use.