AreaAudit

Which languages each generator says it supports, and where

Why one language list is three claims wearing one label

These pages carry the distinctions the atlas is built on. Each exists because a single word on a vendor page is doing the work of three separate claims. As of 2026-09-12.

One word, three promises, and the gap between themSupport is used for menus, for generated output and for voices. The three claims have different costs to make and different consequences if wrong. A page that announces a number of supported languages is nearly always counting interfaces, and is nearly always read as counting voices.What a support claim usually coversOnly interfaceCounted in headlinesMenus, Help centre, Billing screensBoth listsClaimed for bothA short list of major languagesOnly outputNeeded by a productionDialogue the model writes, A voice tospeak itThe middle column is where the two lists overlap
Fig. 1 The number in a headline and the number a production can use are rarely the same number, and no vendor page says which one is on offer.

Support is the word in question. Applied to an interface it means menus. Applied to output it means the model will write or speak the language. Applied to voices it means a specific inventory exists. A page announcing thirty languages is usually making the first claim and being read as making the third.

Another explainer covers blanks. An empty cell in this atlas means the vendor's pages do not address the question, which is not the same as a vendor saying no. The distinction matters most in the markets where coverage is thinnest.

None of these pages puts a language in the register. They are here so that a reader can tell which of the three claims a vendor is making, and can disagree with a specific step rather than with the atlas as a whole.

1The explainers

  • Three layers — The language of the interface, of the documents a tool generates, and of the dialogue it speaks are three separate settings that move independently.
  • How hreflang works — Language alternates tell search engines which URL serves which language.
  • Subtitles and localisation — Script, audio and subtitle are separate layers with separate costs.
  • Shapes of a figure — A language claim can be a named set, a printed total, a bare count or a lower bound.
  • Which variant — Whether a language list means the Brazilian or European form, the Castilian or Latin American form, and why almost nothing published says.
  • Blanks as evidence — What a blank in a coverage table should mean, what it usually gets read as, and why recording absences changes the shape of the whole comparison.
  • Two-sided lists — Why a conversion tool has two language lists, why the shorter one is what a buyer purchases, and how rarely both are published.
  • When the choice is made — Whether a language is chosen before anything is generated or applied to a finished file afterwards, and how much of a schedule that decides.
  • Scripts in pictures — The one language layer that cannot be repaired after a shot exists, which scripts fail soonest, and how to arrange a production around it.
  • Voices for a series — Why a language count is not a cast, what a serialised production needs instead, and which of those requirements anybody publishes.
  • Shapes of a set — Reading a list of site languages as a map of where a company sells, what the ordering reveals, and where the reading stops working.
  • Why no single score — What would be needed to combine three language columns into one number, why no weighting is defensible, and what to do instead.
  • Check one yourself — The whole method this atlas uses for the interface column, written out so a reader can repeat it on any vendor in a few minutes.
  • Regions nobody addressed — Reading a column of declared language sets as a map of where this industry has not gone, and why the gaps cluster by region and by script.
  • The document layer — Why the language of an outline or a breakdown decides whether a crew can work at all, and why almost nobody in this category publishes it.
  • Figures that never expire — Why so many language claims are written as lower bounds, what that does to their value as evidence, and how to read one.

A register entry that depends on one of these distinctions points at this section instead of compressing the case into a parenthesis.

The rest of the atlas: Columns, Entries, By language, Side by side, Markets, Tools, Output, Questions, Terms, Data. How a list earns a column is set out on the counting page.

2How one generator handles languages

On the language side of “Learn”, SceneMixer's published guides say the following:

The interface labels follow your language, and the storyboard text is shown in the project's display language.

SceneMixer, Editing the Storyboard: What You Can Change, and What You Can’t — “Can I write in my own language?” (read 2026-10-13)

A producer working in German can make a series whose cast speaks Indonesian: the interface and the working documents follow the interface language, the spoken dialogue follows the project setting.

SceneMixer, AI Short Drama in 15 Languages (2026): Interface, Scripts and the Dialogue Your Cast Actually Speaks — “Can the interface be in one language and the dialogue in another?” (read 2026-10-13)