Midjourney Tile: Patterns That Repeat with --tile (2026)

TL;DR: --tile makes one square image whose edges line up with copies of itself. Midjourney's Feature Compatibility chart ticks it for V6, V7 and the merged V8.1 and V8.2 column, and the Tile article tells you not to upscale the result. It is starting artwork, not a production pattern file.

What does Midjourney's --tile actually produce?

One image. Not a pattern, not a sheet, not a file with a repeat baked into it.

Midjourney's Tile article is unusually plain about the mechanism: the parameter "makes sure that the edges of your image match up perfectly with the same image when you place them next to each other." The promised result is that "you can use the image over and over, and it will look like one continuous design without any obvious seams or breaks where the images meet." The article's own headline for the parameter is "Create seamless repeating patterns with the tile parameter", and it names the intended uses directly: "It's a great tool for creating designs that need to repeat smoothly, like on fabrics or wallpapers!"

Then it draws the boundary in one clause that most tutorials skip. When you use --tile, "keep in mind that it generates just a single tile."

That distinction runs through everything below. You are commissioning a unit cell; whether it becomes a pattern happens later, in whatever tool you lay it out in.

hand-painted gouache wildflowers, small motifs, cream ground --tile --ar 1:1 --v 8.2 --stylize 150

Midjourney also tells you where to look. Its instruction is to "use a pattern-making tool like a Seamless Pattern Checker to see how multiple tiles fit together", and it documents a built-in route as well: "On midjourney.com you can also click your tile image to open it, and then click it again to see a preview of the full repeating pattern."

Which Midjourney versions support --tile?

All the current ones, which is rarer than it sounds.

The Feature Compatibility chart in Midjourney's Version article has three columns, headed V6, V7, and a single merged column for V8.1 and V8.2. The Tile Parameter row carries a check icon in every one of them. Compare that with the rows either side of it in the same table: Character Reference, Character Weight, Multi-Prompting and the Quality parameter are all marked unsupported in that V8.1 and V8.2 column, and Omni Reference is annotated "(Uses V7)".

The older lines are covered by a second chart, in the Legacy Features article, whose Tile row ticks V6, V5, V3 and Niji 5 and leaves V4, Test, Testp and Niji 4 blank. The V5.1 description in that article lists among its improvements support for "advanced features like repeating patterns with --tile".

Two things follow. The merged column means Midjourney publishes no separate answer for V8.1 and V8.2, so any difference between them is undocumented. And the current default matters: the Version article states that "V8.2 released as the default version on July 24, 2026", so a tile tutorial written against V5 describes a different model. Our Midjourney version flags reference tracks which line is current.

One adjacent gap worth knowing before you try to weight a motif against its background: :: multi-prompting is marked unsupported on V7 and on V8.1 and V8.2 in the same chart. If you were planning to write flowers::2 background::1 --tile, read what Midjourney still documents about prompt weights first.

How do aspect ratio and upscaling interact with --tile?

Upscaling is documented and the news is bad. Aspect ratio is not documented at all.

Start with the sentence that should change your workflow. The Tile article: "Upscaling your tile images will often break the seamless repeating pattern, so we do not recommend upscaling these images." No exception is offered for the Subtle upscaler, which the Upscalers article otherwise describes as the one that "keeps changes to a minimum".

There is a second-order effect that is easy to miss. The only aspect-ratio caveat Midjourney publishes is scoped to upscaling: "Some aspect ratios may be slightly changed when upscaling." For a tile, a proportional change is not cosmetic, because the edges were computed for the original dimensions. Both documented hazards point at the same operation, which argues for treating the generated tile as final. Our --ar reference covers the caps and the no-decimals rule per model line.

On ratio itself, Midjourney publishes nothing specific to tiles. The Tile article opens by asking you to "Imagine you have a square image", and the Aspect Ratio article states that "The default aspect ratio is 1:1", but no page says --tile requires a square, and no page says it refuses a rectangle.

One documented interaction that does work in your favour: the Remix article lists which parameters survive a remix, and tile is on it. "Only parameters that normally influence variations will work while using Remix, including: aspect ratio, no, stop, and tile." The Legacy Features chart agrees, ticking Tile in both its "Affects Initial Images" and "Affects Variations & Remix" columns. So you can iterate on the prompt through Remix without losing the tiling behaviour.

What separates a usable tile from a pretty square?

Five decisions, none of which Midjourney documents, all of which are ordinary surface-design craft. This is the part that decides whether the file is worth anything.

Scale relative to the final application. A tile that looks balanced at 1000 pixels on your monitor may be printed at 6 centimetres across on a shirt or 60 centimetres across on a wall. Decide the physical repeat size first, then choose motif density to match. Small dense motifs survive shrinking; a large single motif does not. Run the same prompt at three densities and judge them at final size.

ditsy floral, tiny motifs, dense even coverage --tile --ar 1:1 --v 8.2 --stylize 100
ditsy floral, medium motifs, balanced spacing --tile --ar 1:1 --v 8.2 --stylize 100
ditsy floral, large motifs, generous negative space --tile --ar 1:1 --v 8.2 --stylize 100

No dominant element, or you have made a grid. This is the most common failure and it has nothing to do with the seam. If one motif is larger, darker or more saturated than everything around it, the eye finds every copy of it and reads the lattice they form. The pattern stops looking like a pattern and starts looking like wallpaper hung badly. The fix is in the prompt: ask for even distribution and rule out a focal point.

tossed botanical sprigs, evenly distributed, no single dominant element --tile --ar 1:1 --v 8.2 --no centrepiece, focal point, medallion

Fewer colours survive production. Screen printing, rotary printing and most embroidery work from a limited number of separated colours, and every extra one adds a screen, a plate or a thread. Gradients and photographic shading are the enemy of that process. Specifying an exact count also makes the file far easier to clean up later, because the colours are already distinct.

folk floral in exactly four flat colours, no gradients, no blending --tile --ar 1:1 --v 8.2 --raw --no gradient, shading, halftone

Non-directional beats directional. A directional repeat has a right way up. That means the fabric can only be cut one way, the wallpaper can only be hung one way, and the print costs more in wasted material. A non-directional repeat, with motifs rotated at varied angles, can be cut and placed in any orientation. It is the more flexible file almost every time.

non-directional tossed motif, elements rotated at varied angles, readable from every side --tile --ar 1:1 --v 8.2 --stylize 150

Hard edges and contrast tile better than soft gradients. A crisp shape either meets its neighbour or it does not, and you can see which in one glance. A soft gradient or a diffuse glow near the boundary produces a mismatch that is faint, wide and much harder to spot, and it is the thing that survives your review and shows up on a two-metre print. Flat vector styling and --raw are the friendly territory here.

flat vector shapes, hard edges, high contrast, no soft gradients --tile --ar 1:1 --v 8.2 --raw --stylize 0

Copy-paste tile prompts by what you are making

Swap the subject, keep the structure. Every block ends in --tile and pins a version so it does not drift when the default changes.

Textile and fabric prints

small-scale ditsy floral, hand-painted gouache, cream ground, five colours --tile --ar 1:1 --v 8.2 --stylize 100
block-printed indigo paisley, uneven ink absorption, natural linen texture --tile --ar 1:1 --v 8.2 --raw
tossed watercolour wildflowers, no directional stems, soft chalky palette --tile --ar 1:1 --v 8.2 --stylize 250 --no grid, border, frame

Wallpaper and wrapping paper

art nouveau vine damask, two-tone sage and bone, flat vector, large repeat --tile --ar 1:1 --v 8.2 --stylize 200
1970s wrapping paper, chunky geometric daisies, four flat colours, offset print grain --tile --ar 1:1 --v 8.2 --raw
chinoiserie birds and blossom, hand-painted, muted teal ground, non-directional --tile --ar 1:1 --v 8.2 --stylize 400 --chaos 10

Surface textures for 3D and game use

weathered cobblestone pavement, top-down orthographic, flat even lighting --tile --ar 1:1 --v 8.2 --raw --no vignette, perspective, highlight
rusted riveted steel plate, top-down, uniform diffuse lighting, albedo only --tile --ar 1:1 --v 8.2 --raw --stylize 0
dry cracked desert mud, orthographic top-down, neutral daylight, even exposure --tile --ar 1:1 --v 8.2 --raw --no shadow, gradient

Backgrounds for web and print

subtle off-white paper fibre texture, very low contrast, barely visible --tile --ar 1:1 --v 8.2 --raw --stylize 0
faint navy dot grid on deep navy, minimal contrast, flat --tile --ar 1:1 --v 8.2 --raw --no shading, gloss
soft embossed linen weave, single colour, near-neutral, screen background --tile --ar 1:1 --v 8.2 --raw --stylize 50

Geometric and tessellating motifs

islamic girih tessellation, twelve-point stars, two colours, flat vector line work --tile --ar 1:1 --v 8.2 --raw --stylize 0
interlocking hexagon lattice, thin gold lines on charcoal, crisp edges --tile --ar 1:1 --v 8.2 --raw
interlocking bird tessellation, black and white, hard edges, no gaps --tile --ar 1:1 --v 8.2 --raw --no gradient, shadow

Organic and botanical repeats

dense monstera and fern foliage, flat cut-paper shapes, three greens --tile --ar 1:1 --v 8.2 --stylize 150
pressed botanical specimens scattered non-directionally, warm ivory ground --tile --ar 1:1 --v 8.2 --stylize 300 --chaos 15
lichen and moss on stone, macro, even lighting, muted greens and greys --tile --ar 1:1 --v 8.2 --raw

Abstract and terrazzo

terrazzo, small chips, warm neutral base, four accent colours, even distribution --tile --ar 1:1 --v 8.2 --raw --stylize 50
memphis confetti shapes, six flat colours, evenly scattered, no clusters --tile --ar 1:1 --v 8.2 --stylize 100
suminagashi marbled ink, indigo on rice paper, continuous flowing lines --tile --ar 1:1 --v 8.2 --stylize 400

Seasonal and themed

scandinavian folk winter motifs, snowflakes and reindeer, two-colour knit chart --tile --ar 1:1 --v 8.2 --raw
tossed pumpkins bats and crescent moons, flat orange black cream, non-directional --tile --ar 1:1 --v 8.2 --stylize 150
spring eggs and sprigs, pastel five-colour palette, evenly tossed --tile --ar 1:1 --v 8.2 --stylize 200 --no basket, scene, horizon

Brand pattern exploration

brand pattern built from one rounded-square mark, two brand colours, generous negative space --tile --ar 1:1 --v 8.2 --raw --stylize 0
packaging pattern, small repeated coffee-bean glyph, cream on deep brown, sparse --tile --ar 1:1 --v 8.2 --raw
brand motif study, thin monoline arcs, one accent on off-white, restrained --tile --ar 1:1 --v 8.2 --raw --sref [YOUR_STYLE_CODE] --sw 200

The reusable template

Save this as a prompt template with the volatile parts marked, so the only thing you edit when the default model moves is the version flag.

[MOTIF], [SCALE], exactly [N] flat colours, non-directional, hard edges --tile --ar 1:1 --v 8.2 --stylize [0-1000] --no gradient, border, frame

And a seed-locked A/B, which is the one comparison worth running deliberately, because stylize changes density and density is the variable that decides whether the repeat reads:

terrazzo chips, warm neutral base --tile --ar 1:1 --v 8.2 --seed 4021 --stylize 50
terrazzo chips, warm neutral base --tile --ar 1:1 --v 8.2 --seed 4021 --stylize 400

How do you verify a tile before you trust it?

By repeating it. Never by looking at the single square.

This is the discipline the whole page rests on, and it is the one Midjourney itself prescribes: use a pattern checker "to see how multiple tiles fit together", or open the image on midjourney.com and click again for the built-in repeat preview. A defect at the join is invisible in one view by definition, because in one view the join is not on screen.

Do it in this order.

1. Look at the built-in preview first. It is free and it catches the obvious failures in a second.
2. Then lay it out at 3x3 in an image editor. Nine copies is the minimum that shows both the horizontal and vertical seams and the diagonal continuity across a corner.
3. Zoom out until the tile is roughly its final physical size. Grid artefacts appear at viewing distance, not at 100 percent.
4. Squint, or blur the layout heavily. Blurring leaves only the tonal structure, which is where a dominant motif or a lighter band at the seam becomes obvious.
5. Check a mirrored copy if the design will be cut and mirrored. Some production layouts flip alternate panels, and a tile can survive a straight repeat and fail a mirrored one.

Then check colour count and edge crispness against the intended process. If the tile needs troubleshooting rather than verification, our image prompt troubleshooting guide covers the wider failure set.

Where does a Midjourney tile stop being enough?

At the point where a supplier asks for a file rather than a picture.

A tiling image is a raster at a fixed pixel size. It carries no repeat metadata: no pitch, no drop, no offset, no repeat width in centimetres. It has no vector paths, so it cannot be scaled without resampling. Its colours are not separated into spot colours, and its edges, however crisp they look, were produced by a generative model rather than drawn to a tolerance.

For textile or wallpaper production this normally means rebuilding: tracing the motifs, rebuilding the repeat properly, reducing the palette to real separations. The generated tile is the artwork and the reference, not the deliverable. Say that to a client before you show them the square, not after.

The print maths is Midjourney's own and it is unforgiving. Its Image Size and Resolution article notes that "printing usually requires around 300 DPI for the best quality", and works the example out: "with an HD 2048 x 2048px image from Midjourney, you can produce a good quality 6.8 inch by 6.8 inch print". For a repeating tile a 6.8 inch cell is often fine. For a tile that must sit alone at scale it is not, and the usual fix is the one the Tile article advises against. GPU cost is worth watching while you iterate, which our quality and speed reference unpicks.

What Midjourney does not publish about --tile

Writing the gaps down beats filling them with plausible answers, and every one is answered confidently somewhere by a page that invented it.

• Whether --tile accepts non-square aspect ratios. No page states a requirement or a restriction.
• Whether --tile behaves differently on V8.1 and V8.2. They share one merged column in the chart.
• Whether Niji 6 or Niji 7 support it. Neither chart carries a row for those models.
• Whether --hd is compatible with --tile. The upscaling warning is about upscaling; HD generation is a different operation and is never discussed on the Tile page.
• What kinds of subject tile well. Midjourney names fabrics and wallpapers and stops there. Every suitability claim you read, including the craft section above, is somebody's experience rather than vendor documentation.
• How often "often" is, in "will often break the seamless repeating pattern". No rate, no condition, no explanation of which upscaler is worse.
• Whether --tile interacts with --sref, --stylize or --chaos. Never discussed.

One legacy interaction is documented, and it has the same shape as the current warning. The Legacy Features article states that "V5 upscale tools are not compatible with the Pan tool or the tile parameter." Tiling and upscaling have been an awkward pair for three model generations.

How to check any of this yourself

Midjourney's parameter surface moved repeatedly through 2026, so verification beats memorisation.

Read the compatibility chart, not the parameter list. The Feature Compatibility chart in the Version article is the only page giving per-version availability for tile, quality, multi-prompting and the rest side by side. Where it disagrees with a topic article, compare update dates. On August 29, 2026 the Version article carried July 24 and the Tile article July 27, while the Parameter List still carried June 11 and still described --hd as a V8.1 feature.

Fetch the docs properly. docs.midjourney.com returns 403 to most automated fetches, which is not a dead link. The Zendesk Help Center JSON API behind it answers plainly at /api/v2/help_center/en-us/articles/<id>.json, and every figure on this page came from there on August 29, 2026.

Test the tile, not the prompt. A prompt that produced a clean repeat last month is not evidence about the tile in front of you now. Lay it out and look.

The durable asset is not any one square. It is the prompt block, parameters included, saved with the volatile parts marked as variables, so that a new brief is a swap rather than a rewrite. That is what we build Prompt Architects for: we generate the prompt, never the image, and the image model stays whichever one you prefer. Image prompt generation starts on the Pro plan per the pricing page comparison table, and the FAQ sets the free plan at five prompt enhancements per day, forever, both checked on August 29, 2026.