QR code generator

How this code was built, step by step

The code above, from text to modules. Change the text or the settings and these steps follow.

The parts of a QR code

A QR code is a square of modules, 21 × 21 at version 1 and 4 more each way per version, up to 177 × 177 at version 40. Some modules are fixed patterns a scanner uses to find and read the code; the rest carry the data and its error correction.

Worked example: 01234567 in numeric mode

The standard's own example, at version 1 and level M, which leaves 16 data codewords and 10 for error correction. The generator above starts with HELLO WORLD in alphanumeric mode, so between them they show both ways of packing characters.

1. Only digits, so the mode indicator is 0001, and 8 digits in 10 bits is 0000001000.
2. Digits go in threes, each three written as a 10-bit number: 012 12 0000001100345 345 010101100167 67 1000011 67 is left over as a pair, so it takes 7 bits. Three digits in 10 bits is about 3.3 bits a digit, against 8 in byte mode.
3. That is 41 bits. 4 terminator zeros, 3 more to finish the byte, and 10 pad bytes make the 16 data codewords: 10 20 0C 56 61 80 EC 11 EC 11 EC 11 EC 11 EC 11
4. Reed–Solomon over GF(256) gives the 10 error correction codewords: A5 24 D4 C1 ED 36 C7 87 2C 55
5. The 26 codewords fill the 208 free modules exactly; version 1 has no remainder bits. Mask 0 scores lowest here, and the format information for level M and mask 0 is 101010000010010. Level M is 00 and mask 0 is 000, so the five data bits and their BCH bits are all zero, and what is left is the fixed pattern the format bits are XORed with.

Open 01234567 in the generator. The codewords are bytes like any other; the hex to binary converter turns them back into the bits in the modules.

Error correction levels

Reed–Solomon codewords are added to every block. A scanner can rebuild damaged codewords from them, up to the share each level is designed for. The cost is room for data.

The format bits are not in alphabetical order: the standard numbers the levels M = 00, L = 01, H = 10, Q = 11.

The eight masks

Data can come out in stripes or blotches, or contain something that looks like a finder pattern. So the data and error correction modules are XORed with one of eight fixed patterns, flipping the modules where the pattern is dark, and the pattern that leaves the fewest problems is kept. Row and column count from 0 at the top left.

The penalty rules, from ISO/IEC 18004:

• Runs. Five or more modules of one colour in a row or column score 3, plus 1 for each module past five.
• Boxes. Each 2 × 2 block of one colour scores 3. Blocks can overlap.
• Finder-like patterns. Dark, light, dark, light, dark in the ratio 1:1:3:1:1 with four light modules before or after it scores 40. A pattern with four light modules on both sides counts twice, 80. This page counts the quiet zone as light, so the three real finder patterns always score 720 between them; only the rest differs from mask to mask.
• Balance. 10 for every whole 5% the share of dark modules is away from half.

The rules leave room for interpretation, at the edges especially, so two correct generators can choose different masks for the same text. Both codes scan, because the mask number is in the format information.

QR code capacity by version

How much fits in each version: bytes at each level (byte mode, such as a link with lower case letters), and digits and alphanumeric characters at level M.

A character outside ASCII takes 2 to 4 bytes in UTF-8, so text with accents or emoji fits fewer characters than the byte count. The binary translator shows the UTF-8 bytes of any text, and the ASCII table the one-byte characters. This generator writes UTF-8 with no ECI header; without one the standard's default for byte mode is ISO-8859-1, but most scanners recognise UTF-8 by its byte patterns.

Common mistakes

• No quiet zone. Printing the code flush against a border, a photo or other text. Leave four modules of light margin on every side.
• Too dense to scan. A long link makes a high version with small modules. Shorten the link, or print the code bigger.
• Lower case in a link that could be capitals. One lower case letter forces byte mode. At level Q, https://logicgates.org needs version 3; HTTPS://LOGICGATES.ORG fits version 2. Scheme and domain are not case sensitive, but the path after them can be.
• Low contrast or inverted colours. Dark modules on a light background is the normal form and the safest. Many phone scanners also read light on dark, but not all.
• A logo over too much of it. A logo destroys the modules under it, which error correction has to make up for. Use level H, keep the logo small and away from the finder patterns, and test the result.