QR Tools

Every tool here produces or reads the same object: a grid of dark and light modules that decodes to a string of characters. ISO/IEC 18004 defines that grid. It specifies the three finder patterns, the alignment patterns, the four encoding modes (numeric, alphanumeric, byte, kanji), the versions running from 21x21 modules up to 177x177, and the Reed-Solomon parity that lets a scratched symbol still resolve. What the standard does not define is meaning.

The payload is plain text, and most of its conventions are informal

A scanner hands the decoded string to the operating system, which guesses what to do with it. A leading https:// opens a browser. A leading tel: dials. A string that begins WIFI:T:WPA;S:Backroom;P:swordfish;; joins a network, but only because Android's ZXing library settled on that format and everyone copied it. No standards body ratified it. That is why the escaping rules bite: a semicolon, colon, comma or backslash inside an SSID or password must be prefixed with a backslash, and an all-digit password sometimes has to be quoted so it is not read as hex. When a phone refuses to join, the fault is nearly always in the string rather than the symbol, which is what a WiFi QR decoder is for: paste the raw text, read the fields back, find the unescaped character. The WiFi generator writes that string for you, the decoder and the screenshot scanner read it back.

Error correction, and why the highest level is often the wrong choice

There are four levels. L recovers roughly 7% of the codewords, M about 15%, Q about 25%, H about 30%. Parity costs capacity, so raising the level either reduces what you can encode or forces a larger version. Take a 60-character URL in byte mode. At level L it fits version 4, a symbol 33 modules square. The same 60 characters at level H needs version 7, which is 45 modules square. Print both at the same width and every module in the H version is about 27% narrower, which is harder to scan.

So fix the payload before you touch the parity. Trim that URL to 25 characters and version 4 holds it at level H: back to 33 modules, with the strongest correction available, at no cost in size. Case helps as well. Alphanumeric mode packs two characters into 11 bits against 8 bits for byte mode, but its 45-character set has no lowercase, so HTTPS://EXAMPLE.COM encodes smaller than the same host in lowercase. Hostnames are case-insensitive; paths usually are not.

Choose the level by surface. L suits a screen or a page inside a book. M is the sensible default for clean paper. Q or H belong on anything that gets handled, curved, rained on, or partly obscured: a van door, a laminated plant tag, a code with a logo dropped into the middle. A centre logo is damage you inflict deliberately, it has to stay inside the recoverable fraction, and it must never encroach on the finder patterns.

The quiet zone is four modules

Four modules of blank margin on all four sides. Not two, not whatever padding looks tidy. Decoders locate a symbol by the 1:1:3:1:1 run ratio through each finder pattern and need clear space to know where the code stops. Label failures are usually this: someone cropped tight to the black, or floated the code over a photograph. A transparent background makes the risk worse, because contrast then depends on whatever sits behind it in the final layout. Check it after placement.

Print cannot be edited

Once ink hits stock the destination is frozen. Encode a URL on a domain you control and redirect from there, never a vendor's tracking link and never a deep path into a platform that reorganises its URLs. For volume work, generating from a spreadsheet keeps one row mapped to one symbol, which is the only practical way to audit 300 asset tags before printing.

Unknown codes deserve the same care. A scanner that shows the decoded string first lets you see the full destination before a browser acts on it, and screenshot decoding covers codes you cannot point a camera at, such as one inside a PDF invoice.