Entries that answer with a number and no names
Three entries answer with a number. Five, thirty or more, and ninety-five or more. The largest belongs to a model that works from a waveform and so has no per-language voice inventory to publish in the first place. As of 2026-09-22.
| Model | The published total | And where the sound is made |
|---|---|---|
| HeyGen | A count, attached to translation | From a speech endpoint, then rendered |
| Kling AI | A count, no names | With the picture, on VIDEO 3.0 |
| sync-3 | A count, ninety-five or more | Supplied, or read from text |
Inclusion rule. Entries whose documentation gives a number of spoken languages without naming them. An unnamed claim of breadth with no number is on a separate route page. Order. Alphabetical by model name.
1A large total can mean less work rather than more
Ninety-five languages sounds like an enormous inventory and probably describes the opposite. A model animating a mouth to supplied audio needs no voice per language; it needs alignment to hold across the sounds those languages make. That is a claim about phonetics, not about casting.
Read that way the figure becomes more credible and less useful. Credible because alignment generalises; less useful because it says nothing about whether a production can obtain a voice in the language it needs.
2Where a total sits tells you what it is for
One of these attaches its figure to video translation, which describes a workflow: one master, one performance, and a localisation stage at the end. Another attaches it to native audio on a specific model version, which makes the capability a property of a version rather than of an account.
Neither figure transfers to the other arrangement. A count that belongs to translation says nothing about generating in a language from the start, and a reader comparing the two numbers side by side is comparing different things.
3The smallest total is the one that hurts most
With a large number a production can assume its market is probably included and check cheaply. With five, the probability runs the other way, and an assumption becomes a commitment made on a coin toss.
Five is also short enough that naming the five would have been decisive, which is why the omission reads less like an editorial decision about page length and more like a capability still settling.
4The entries on this route, one page each
Each of these links to that entry on the field this route groups by. The wording behind every cell, and the date it was read, sits on the page it links to.
- HeyGen — a count, attached to translation.
- Kling AI — a count, no names.
- sync-3 — a count, ninety-five or more.
- Languagessync-3 supports 95 or more languages, described as the same coverage as the models before itpublished as a count
- LanguagesVideo translation is documented for thirty or more languages, with voice cloning and lip-synccounted on the translation endpoint
- Audio sourceNative audio on VIDEO 3.0 in five languagesgenerated with the picture
- Audio sourceThe accepted pairs are video with audio, video with text, image with audio and image with textsupplied or read from text
5Sources
Membership of this route is decided by the wording in the field it groups by, read from the documentation each vendor publishes. The column itself is on the field note, all five columns are on the speech table, and what counts as documented is on how read. Other routes: Breadth claimed, A figure in writing.