AreaAudit

Which languages each generator says it supports, and where

Two vendors whose answers point opposite ways

A pairing is only worth a page when the two vendors answer the same column differently enough that the difference teaches something. Two vendors that both publish nothing are not a comparison. As of 2026-09-12.

The three shapes of disagreement worth a pairingTwo vendors are worth reading together when they hold the two ends of one column, when they fill it in incompatible ways, or when each is strong in a column the other leaves empty. Two vendors who both published nothing are not a comparison.Largest disagreement firstOpposite endsWidest declared list against narrowest, inside one column.Incompatible fillingA named list against a bare count, or two lists cut alongdifferent axes.Different strengthsEach vendor fills a column the other leaves blank, so the winnerdepends on the question.No pairing is scored or ranked
Fig. 1 Each pairing page states which of these three shapes it is, so a reader can tell what kind of finding to expect before reading.

The useful pairs fall into three shapes. Two vendors at opposite ends of one column; two vendors that fill the same column in incompatible ways; and two vendors whose strengths sit in different columns, so that whichever one looks better depends entirely on the question asked.

Nothing here is scored. Each page prints what both vendors published, names the one thing that separates them, and says which reader would care. A pairing where the honest answer is that both rows say the same thing does not get written.

Both halves of every pairing are also read on their own, cell by cell, so a reader who wants one vendor rather than a contrast can leave the comparison at any point.

1The pairings

  • CapCut and D-ID — The widest and the narrowest declared site lists in this register, and why the gap between them predicts nothing about what either product can generate.
  • CapCut and Fliki — One vendor has translated its site into 27 locales and names no voices; the other declares no site language and counts 91 of them.
  • Synthesia and Kling AI — The longest named language list in this register against the smallest published count, and why the gap is about checkability rather than size.
  • Synthesia and HeyGen — Two vendors publishing named language lists, one countable at a glance and one only by hand, with marketing figures above both that disagree.
  • VEED and HeyGen — Both publish two spoken-language lists, but one divides input from output and the other divides voices from finished-video translation.
  • Rask AI and Vidnoz — The two vendors here whose declared alternates sit on language subdomains, and the one of them that also carries a code pointing at the wrong language.
  • SceneMixer and Synthesia — The one vendor here publishing in every column, set against the vendor with the longest named list and nothing at all about generated documents.
  • D-ID and Rask AI — Two large spoken-language counts with no set behind either of them, published by two companies sitting at opposite ends of the site column.
  • Kling AI and PixVerse — Two bare-code declarations that both reach Southeast Asia, where they agree, and diverge everywhere the smaller set stops.
  • Runway and Vidu — Two site-only vendors whose declared sets differ in size, in the form of the codes, and in whether German appears in either of them at all.
  • Higgsfield and D-ID — The two shortest declared sets in this register, one contained in the other, and the one of the pair with a page that was opened and read.
  • Fliki and HeyGen — Two vendors with no site alternates and long spoken-language lists, differing only in whether the company did its own counting for it.
  • CapCut and Synthesia — Two declared sets that differ in form as much as in size: one names a market for every language, the other for one of nine.
  • VEED and SceneMixer — Two published mechanisms for getting a video into another language, and the different questions each of them leaves a producer with.
  • Vidnoz and Fliki — Both vendors publish a pair of language figures.
  • Kling AI and SceneMixer — Two vendors whose site columns are close and whose voice columns are not, and the Chinese alternate both of them lose to the same rule.
  • Runway and Higgsfield — The two declared sets here of identical size, which overlap on three codes and disagree about German, French and Spanish entirely.
  • PixVerse and Vidu — The one exact containment between two declared sets here: nine codes inside thirteen with nothing dropped, and four Southeast Asian additions.
  • Synthesia and Fliki — The two longest named lists in this register, one totalling languages and dialects together and one keeping them in separate counted sections.
  • HeyGen and Rask AI — Two vendors whose spoken-language disclosure is the exact complement of the other: one publishes the set, the other only the size.

Pairings are chosen for the size of the disagreement, never to fill a grid of every vendor against every other.

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