Zalgo text generator

Zalgo started on the Something Awful forums in 2004, where a member posted image macros of cartoon characters, distorted and glitched, crying “Zalgo!”. Writing made to look the same way, with letters that seem to melt into the lines around them, took the name too. It has been a fixture of creepypasta horror stories ever since, and later of surreal memes.

Programmers know it from a famous 2009 Stack Overflow answer about why HTML can’t be parsed with regular expressions, which slowly dissolves into Zalgo as it goes on.

Real languages stack marks too: the Vietnamese ệ carries a circumflex above and a dot below. Unicode handles this with combining marks, characters that have no width and attach to whatever comes before them, and it doesn’t limit how many one letter may carry. The nearest thing to a cap is the “stream-safe” format in Unicode’s normalization rules, which software can use when it needs a limit; it allows 30 marks in a row.

This generator takes its marks from the Combining Diacritical Marks block, U+0300 to U+036F. Some are everyday accents, such as the grave and the tilde. Others are small letters that medieval scribes wrote above other letters, such as ͣ and ͤ. In the heavy style each letter gets between one and six marks above and as many below, so a single letter can carry 12, and the font has to keep placing each one further from the line.

All those marks count as characters. “do not look back” is 16 characters as typed and 98 as Zalgo, and a text box with a length limit may count every one of them, so Zalgo can run out of room long before it looks long.

• It spills. Tall stacks run over the comments, messages or lines above and below, covering other people’s words.
• It shuts out listeners. A screen reader may read every mark, skip the word or spell it out, so a Zalgo comment is lost on anyone using one.
• It hides from search. Someone typing the word is unlikely to find it.
• It has been used to cause trouble. Very heavy Zalgo has crashed some versions of Apple’s Messages app, and people have sent it on purpose to do that.

None of that makes it off limits. On your own Halloween post or horror story it does exactly what it should. Just keep it out of other people’s threads and inboxes.

Because each mark is its own character, cleaning Zalgo means deleting the marks and keeping the letters. For a few words, retyping is quickest. For more, any tool with Unicode-aware find-and-replace can strip them: in JavaScript, text.replace(/\p{M}/gu, '') removes every combining mark in one go.

Two catches. That also removes accents typed as separate marks, so check names and words in other languages afterwards. And if the text has passed through an app that merges a letter and its mark into one accented character, such as ẻ, the replace won’t touch it. Running text.normalize('NFD') first splits those apart, at the cost of also stripping real accents, like the one in café.