Measure before you fix
Seamless Texture Checker
Find out whether a texture really tiles, and which of its two axes is broken, before it ends up in a material. Measured in your browser against the texture's own internal detail.
Load an image to see whether its left edge matches its right edge, and its top edge its bottom.
Need it fixed instead of measured? Run it through the seamless texture maker — the blend removes both seams in one pass.
Nothing is uploaded: the image is read with the Canvas API on your device. Useful for vetting downloaded texture packs and for double-checking your own exports.
What the score measures
The edge mismatch between left and right, and between top and bottom, divided by the average difference between neighbouring pixels inside the image. A score of 95 means the join is as smooth as the texture's own interior.
Every texture that fails in a render looks fine on its own. The break only appears when the surface is large enough to need a second copy, and by then the material is already built, the UVs are already unwrapped, and the fix costs an afternoon. A measurement costs one second.
This checker compares the two pairs of edges that decide whether a texture tiles: left against right, and top against bottom. Then it compares that mismatch with how much neighbouring pixels differ inside the image, because the question is not whether the edges differ — every texture differs pixel to pixel — but whether they differ more than the texture normally does.
Reading the score
90–100: it already tiles. The edges are as smooth as the interior. Stop blending; a blend here would only soften detail you wanted to keep.
60–89: one axis is off. One of the two readouts is clean and the other is not. That is the signature of an image that was cropped from a larger seamless source along one direction — common with texture packs that ship strips. Fix the broken axis and leave the other alone.
Below 60: neither axis tiles. Both joins are sharp discontinuities. This is what a raw photograph scores, and it is normal: nothing is wrong with the image, it just was not built to repeat.
Why the numbers can be trusted
A percentage on its own says nothing. A mismatch of 12 units between edges is invisible on a heavily noisy texture and glaring on a flat plaster wall, because noise hides discontinuities while flatness exposes them. Dividing the edge mismatch by the average internal difference makes the score relative to the texture itself, so a noisy gravel scan and a smooth painted wall can both score 95 and both really do tile.
The preview is the second half of the answer. It repeats your texture the way the final surface will, and the dashed box marks the tile boundary — the place where a repeat line would appear. If the score says 95 and you can still see a line in the preview, trust your eyes: the line is real, and it usually comes from a lighting gradient or a shadow that survived the crop.
What causes a seam when the edges do match
The frustrating case is a texture whose left and right edges agree perfectly and which still shows a repeat. Three usual culprits:
- A vignette or uneven lighting. A photo lit from one side is darker at one edge than the other. Even after the edges are matched numerically, the gradient persists and repeats as a visible band.
- A low-frequency colour cast. The average colour drifts across the frame, so the tiles are individually fine and collectively striped.
- A recognizable feature. A single leaf, scratch or stain will be spotted the moment it appears twice.
All three are fixed before the seam: flatten the lighting, equalise the average colour, and crop away anything the eye can name. The blend cannot fix a gradient it was not told about.
From measured to fixed
Once you know which axis failed, the maker handles it: it offsets the image so the broken join moves to the middle, then blends both seams with an adjustable width. Run the result back through this checker — the score should climb above 90 — and export only then.
Keep the numbers with the texture. A short note in your library (“brick_red_02, checker 96, blend 12%”) saves the next person from repeating the measurement, and it saves you when the same material comes back in a year.
Checker questions
How do I check if a texture is seamless?
Load the image and read the two edge rows. If one reads clean and the other does not, the texture was cropped along one direction from a larger seamless source: fix only the broken axis. If both are dirty, treat it as a raw photograph and run it through the maker.
Can a texture score well and still show a seam?
It can be. A perfect score with a visible line in the preview almost always means a lighting gradient or vignette in the source. The edges match numerically but the brightness drifts across the frame, and the drift repeats. Flatten the lighting before blending.
Can I hide a seam with UV tricks instead?
No. Each repeat line would fall on a different row or column, and the UVs would have to be re-authored per tile. Making the source image itself tileable is the only version that survives real use.
Why are horizontal and vertical measured separately?
Both are measured separately, and the lower of the two decides the score. A texture that only tiles horizontally is still not usable on a wall, and reporting a single blended number would hide that.
Does the checker work on normal maps too?
Yes, and it uses the same maths as this page: the edge pair is compared with the interior difference. Expect a low score from a photograph and a high score from a procedural pattern generated with wrapping noise.