The label proofed perfectly. Flat on the desk, printed on paper, the code read from half a meter every time. Wrapped around a slim can it reads from one narrow angle, only up close, and on the shelf beside eleven other cans it does not read at all. The artwork never changed. The surface did.
Curvature breaks the grid the decoder expects
A decoder finds the three finder patterns, works out the transform that maps the printed symbol back to a square grid, then samples the center of each module on that grid. The math assumes a flat plane viewed at an angle.
A cylinder is not a flat plane. As the code wraps, module spacing compresses toward the left and right edges, and it compresses non-linearly, so a single perspective transform cannot describe it. Sampling drifts off the module centers near the edges, and the further it drifts the more bits come back wrong. Error correction absorbs some of that, then it runs out.
The wider the code sits relative to the object's diameter, the worse the compression gets. The same code on a five-liter jug behaves far better than on a lip balm tube, because the arc it covers is flatter.
Alignment patterns help, and then they stop helping
From version 2 upward the symbol carries alignment patterns, small squares distributed inside the data area. They exist so a decoder can resample locally instead of trusting one global transform across the whole symbol, which is exactly what mild warp needs. The anatomy of finder, alignment and timing patterns explains where they sit and why.
Their limit is spacing. They are far apart relative to a module, so they correct gentle, gradual distortion. They do not follow a tight radius, a fold or a sharp change in curvature, and a code that crosses a bottle shoulder or a can crimp goes past what they can fix.
Keep the code inside the flattest arc
Practical placement, in the order that matters:
- Keep the wrap small. Hold the finished object at arm's length. If the left and right edge modules already look foreshortened from a straight-on view, the code is too wide for that diameter and needs to shrink in width or grow in diameter.
- Never cross a seam. Label overlaps, shrink-sleeve seams, crimps, welds and handles all break the surface, and a break anywhere in the symbol is worse than the same amount of area lost at the margin.
- Avoid compound curves. A single-axis cylinder is recoverable. A bottle shoulder, a sphere or a molded corner curves in two directions at once, which nothing in the format was designed to handle.
- Cut the payload, not the size. Fewer characters means a lower version, fewer modules and physically larger modules at the same footprint. A short redirect URL is the single most effective change available for small containers.
- Then go up in size. Work from the scan distance in the size guide and treat curvature as a reason to exceed it, not meet it.
Shrink sleeves add one more variable. The film distorts as it shrinks, unevenly, so artwork is usually pre-distorted in prepress. How much and how well depends on the converter and the container. Ask for a printed and shrunk sample on the real container before approving anything.
Fabric: the weave sets the smallest module you can print
On cloth the limiting factor is not your file. It is the structure of the material and the print method together.
Thread pitch and surface texture decide how cleanly a module edge can land. Ink wicks along fibers, so edges go ragged, and on a coarse or fuzzy surface a small module loses its shape entirely. Tightly woven cotton poplin, canvas and twill hold detail. Fleece, terry, ribbed knit and heavy jersey do not.
Print method matters as much:
- Screen print and transfer on tight woven fabric give the crispest edges.
- Direct-to-garment on a soft knit is softer and needs bigger modules.
- Woven labels hold surprisingly fine detail for their size.
- Embroidery works in stitches, and stitch pitch is coarse compared with a module. Small codes are effectively out of reach there; treat embroidery as unsuitable unless the code is very large and very low version.
On dark garments, print a light block with a generous margin and put the dark code inside it. Do not invert dark and light and hope, and do not let the garment color serve as the light modules.
Stretch and wash are separate problems
Stretch distorts the grid while the garment is being worn, which is exactly when someone tries to scan it. Put the code on a panel that does not move: an upper back, a tote panel, a flat pocket area on woven fabric. Keep it off chest fronts on knit tees, sleeves, cuffs, ribbing, waistbands and caps, which curve and stretch at once.
Washing is cumulative damage. Prints crack, fade and lose contrast, and a code that read on day one may not read after twenty cycles. Assume degradation and build margin in before the run.
That margin is where error correction gets misread as a free upgrade. Level H tolerates roughly 30 percent damage against roughly 7 percent at level L, but the redundancy is stored as extra modules, so at a fixed physical size level H gives you smaller modules. On coarse fabric, smaller modules can hurt more than the extra tolerance helps. Shorten the payload first, size up second, and only then raise the level, using the comparison in error correction levels.
Test on the object, not the flat proof
Order a real sample and treat it as a pass-or-fail gate. Apply one label to the actual container and print one garment on the actual fabric. Wash the garment three times. Then scan both at the distance and angle a customer will use, on at least two phones, on a shelf or a hanger rather than a desk.
Generate the code with a short redirect and heavier module weight in the custom QR code generator, send that sample request to the converter this week, and hold the artwork until the scanned sample comes back clean. If the sample fails and the cause is not obvious, the troubleshooting checklist and the print setup notes will narrow it down faster than another proof round.
