One vendor and one column, a cell at a time
The register table gives every vendor one row and three cells. These pages take a single cell and read it closely: the codes, the form they are written in, the count beside them, and the question the cell still leaves open. As of 2026-09-12.
A cell is the smallest thing this atlas records, and it is where the disagreements live. Two vendors can both have a filled site-language cell while one has named 27 markets and the other has named a language without naming a market at all.
Cells get a page when there is something in them worth a close reading, or when the silence in them is strange next to what the same vendor publishes elsewhere. A vendor that publishes nothing anywhere has nothing to read closely and gets no cell page.
Every cell page carries the column it belongs to beside it, so the thing being read is never mistaken for the whole row. The row itself is on the tool page, and the table with every row in it is the register.
1The cells
- CapCut: site language — The longest declared site list in this register, written almost entirely as language-region pairs, and the six languages no other vendor declares at all.
- D-ID: site language — The shortest declared site list in this register, and the one set that sits inside every other declared set here without adding anything to it.
- Fliki: site language — A vendor whose own pages declare no language alternates at all while publishing the most precisely counted spoken-language list in this register.
- HeyGen: site language — No declared alternates on the host, while the help centre that holds the language lists is itself served in one language only.
- Higgsfield: site language — Five declared alternates, and the result of opening one of them to see whether a declaration corresponds to a page actually written in that language.
- Kling AI: site language — The second-longest declared site list here, written entirely as bare codes, and the one set here that names Hindi or Filipino.
- PixVerse: site language — Thirteen bare codes, nine of which are precisely the set another vendor in this register declares, plus four Southeast Asian additions.
- Rask AI: site language — The largest gap in this register between what a vendor declares and what the counting rule keeps, plus one code that does not match the page it points at.
- Runway: site language — A short declared set with no German in it, and the one mixed-form set here: four bare codes beside a single region-tagged Brazilian Portuguese.
- SceneMixer: site language — Fifteen declared alternates, one of them tagged by writing system rather than country, and a sixteenth for Chinese that points off the host.
- Synthesia: site language — The one declared set here that names Swedish while leaving out Korean, and one of the three that pick the Brazilian variant of Portuguese.
- VEED: site language — An empty site cell beside the one published list in this register that separates the languages a tool can hear from the ones it can speak.
- Vidnoz: site language — Twelve language alternates declared and eleven served from language subdomains, which the counting rule discards as it does everywhere else.
- Vidu: site language — Nine declared codes, every one of which also appears in a thirteen-code set here, and a set that reaches East Asia without reaching Southeast Asia.
- CapCut: dialogue language — An empty voice cell at the vendor with the longest declared site list, and the subtitle figure on its own pages that is routinely read as a voice figure.
- D-ID: dialogue language — A large spoken-language claim with no list behind it, on a vendor whose own site is declared in two languages, and what a reader can do with that.
- Fliki: dialogue language — The one vendor here whose list page prints its own totals, and whose marketing figures sit below the list its own pages publish.
- HeyGen: dialogue language — Two alphabetical help-centre lists with no count on either, a marketing figure of 175+, and the one per-job language limit published in this register.
- Kling AI: dialogue language — The smallest published spoken-language figure here, attached to a model guide rather than a marketing page, and still without a list.
- PixVerse: dialogue language — An empty voice cell beside a launch-post claim about multilingual text drawn inside the picture, which is a fourth kind of language claim.
- Rask AI: dialogue language — The second-largest spoken-language count in this register with no list behind it, published on the home page of a vendor built around dubbing.
- SceneMixer: dialogue language — A named dialogue set matching the interface list, one extra language available for speech only, and a stated method with no separate dubbing pass.
- Synthesia: dialogue language — The longest named language list in this register, the two different counts printed above it, and the one language supplying fourteen of its rows.
- VEED: dialogue language — The one published list in this register with an input side and an output side, and the reason the shorter of the two numbers is what matters.
- Vidnoz: dialogue language — Two spoken-language figures inside one document with nothing distinguishing them, and why the register prints both instead of choosing.
- SceneMixer: document language — One vendor states what language its generated scripts and breakdowns come out in.
A cell earns a page when its contents, or its emptiness, would surprise a reader who had only seen the totals.
The rest of the atlas: Columns, By language, Side by side, Markets, Tools, Output, Questions, Terms, Learn, Data. How a list earns a column is set out on the counting page.