AreaAudit

Which languages each generator says it supports, and where

Diacritic: the marks that change a word rather than decorate it

A diacritic is a mark added to a letter. In several languages it distinguishes words rather than decorating them, so dropping one is not a typographic slip but a spelling error, and generated pictures drop them constantly. As of 2026-09-22.

Where a dropped mark is a spelling error and where it is nothingOn a page or in a subtitle a missing mark is a wrong word and any reader sees it. In a generated picture it is the first thing to disappear. In spoken dialogue it is irrelevant, because the distinction the mark records is audible without it.Text layersSpoken dialogueA dropped markA different word, or no wordNothing, the mark is not audibleWho noticesAny reader of the languageNobodyFixable laterIn a file, if the file is separateNot applicableIn a pictureNot fixable at allNot applicableMarks are a property of writing rather than of speech
Fig. 1 That asymmetry is why text claims and voice claims belong in different columns: one language can be fragile in one and easy in the other.
Where a dropped mark does and does not matter. Recorded 2026-09-22.
ContextEffect of dropping a markNoticed by
A marketing pageA visible spelling errorAny reader
A subtitleA wrong word, or an unreadable oneAny reader
Text drawn in a pictureA word that does not existAny reader
Spoken dialogueNothing, the mark is not audibleNobody

Inclusion rule. The places a diacritic appears in a production, with what happens when it is lost. Spoken output is unaffected because the mark is orthographic. Order. Fixed order, text contexts first.

1In some languages the mark is the word

Several languages read here use marks to distinguish vowels, tones or consonants, sometimes two marks on one letter. Dropping one produces a different word or no word, which a reader notices immediately.

That makes these languages a good test of whether a system is rendering text or approximating it. A page or a subtitle with the marks intact has been through something that understands the script.

2Pictures lose them first

A model drawing letters inside a frame works from shapes. Marks are small, they sit outside the letter body, and they are the first thing to disappear or to land in the wrong place.

One vendor here claims multilingual text generation inside frames and names two languages before saying others. Neither named language is diacritic-heavy, and the claim cannot be read as covering or excluding one that is.

3Nothing is lost in speech

A diacritic is a property of writing. A voice speaking the language produces the distinction the mark records without needing the mark, so a voice inventory is indifferent to it.

That asymmetry is a small illustration of why this register keeps text claims and voice claims in separate columns. The same language is easy in one and fragile in the other, and a merged figure would hide which.

A definition, not a measurement: no vendor figure appears here that is not also in the register. The diacritic-heavy languages declared here are listed on their own language pages.

Nearby terms: Right-to-left text, Tonal language. The whole vocabulary is at terms; nothing on this page is a claim about a named product.