Make a QR code worth printing.

Every QR code carries redundant data so it still reads when part of it is damaged. There are four levels: L recovers about 7% of the code, M about 15%, Q about 25% and H about 30%. The redundancy is real data, so a higher level needs more modules — the same content at H produces a denser code than at L, and a denser code has to be printed larger.

M is the sensible default for anything on a screen. Go to Q or H when the code will be printed, handled, or covered by a logo. This tool raises the level for free whenever your content leaves spare capacity in the version it picked, so you often get Q without paying anything for it.

The common advice — keep it under 30% at level H — is wrong in both directions, and it is worth understanding why.

Error correction is applied per Reed-Solomon block, not across the code as a whole. Damage concentrated in one block can destroy a code covering far less than 30%, while the same amount of damage spread evenly survives comfortably. A percentage cannot tell those apart.

Worse, a single module of a finder pattern — the three large squares in the corners — is fatal at any error correction level. Scanners use those to locate and orient the code before error correction ever runs. No amount of redundancy recovers a finder pattern.

So this tool does not estimate. It maps the modules your logo covers onto the specific codewords they belong to, works out how much damage lands in each block, and compares that against the block’s actual correction budget. That is the number shown as “error budget used”. Keep it under 70% and you still have room for a scratch or a bad print.

The working rule is that a code reads from roughly ten times its own width, so a code scanned from 30cm wants to be about 3cm across. That alone is not enough, though, and it is the reason most printed codes fail.

You also have to check the module pitch — the size of one individual square. Below about 0.4mm, ink spread closes the gaps between modules whatever the overall dimensions say. A dense code sized correctly for its scan distance can still have modules under that floor, and it will not read.

The size shown in this tool takes the larger of the two constraints, so it never recommends something that satisfies the distance rule but falls under the printable floor.

Low contrast is the most common cause. A scanner only has to separate two populations of pixels, so it needs less contrast than text does — but below roughly 3:1 camera noise and uneven lighting start merging them. Aim for 7:1 or better, and remember that a gradient is only as readable as its palest stop.

Missing quiet zone is the second. The specification requires four modules of clear space on every side. Codes placed flush against artwork lose that margin and become hard to find against a busy background.

Decorative shapes are the third, and the least understood. ZXing-family decoders verify a finder pattern by checking its 1:1:3:1:1 ratio horizontally, vertically and diagonally. A circular eye centre measures 3.0 across its diagonal where a square measures 4.24, so the diagonal check fails. Newer phone cameras use machine-learning detection and tolerate it, which is why circular eyes look fine on a modern handset and fail on older scanners and cheap hardware. Every shape in this tool is labelled with the margin it costs.

Finally, inversion. Light modules on a dark background scan on phones, but a meaningful minority of point-of-sale and industrial readers only look for dark on light.

Every code this tool makes is static: the content is encoded directly into the pattern. It works forever, offline, with nothing to expire and no one counting your scans.

The trade-off is that the content cannot be changed after the fact. Services that offer editable codes do it by encoding a link to their own server and redirecting — which means the code stops working if they go away or you stop paying, and every scan passes through them first.

If you need to change a destination later, point the code at a URL you control and change what sits behind it. You keep the ability to edit, and nobody else sits in the middle.