Five spoken languages, and the smallest count in the register
Kling AI states in its model guide that its 3.0 video model produces native audio in five languages, with voices bound to elements. Which five is not stated, so this is the smallest count in the register and still not a set. As of 2026-09-22.
| Vendor | Count published | List published |
|---|---|---|
| D-ID | More than 120 languages and accents | No |
| Kling AI | Five, on the 3.0 video model | No |
| Rask AI | More than 135 languages | No |
| Vidnoz | Both 100+ and 140+, on one page | No |
Inclusion rule. Every cell in this column where the vendor published a number and no list. Vendors publishing a named list are excluded. Order. Alphabetical by vendor.
1A small count is not a worse claim, only a smaller one
Five against 120 or 135 looks like a weak entry until the unit is checked. This figure is attached to one named model and to audio generated as the shot is produced, which is a narrower and more specific claim than a company-wide inventory total.
Narrow claims are easier to keep true. What it still does not do is name the five, so a producer asking whether Portuguese is among them has no more of an answer here than at the vendor claiming 135.
2Voices bound to elements is a second fact
The same sentence states that voices are bound to elements, which is a statement about how a voice is assigned within a generation rather than about how many exist. It is the only description of that mechanism in the register.
For a series, assignment matters as much as inventory: a voice that cannot be pinned to a character drifts between episodes. The register records the wording without treating it as a promise about consistency, which the page does not make.
3Where a count sits changes what it is worth
This count is on a model guide. The other three are on marketing or product pages. A figure in documentation is usually maintained alongside the thing it describes, which makes it more likely to be current than a landing-page figure.
That is a reason to prefer it, not a reason to trust it more than a list. The cell reads five, count only, and it will read more when the vendor names them.
4Sources
Read from the Kling AI 3.0 model guide. The column itself is described on dialogue language, and every cell Kling AI fills is on its tool page.
5The same column, tool by tool
| Tool | Voice languages |
|---|---|
| CapCut | No list published |
| D-ID | 120+, count only |
| Fliki | 91 named, plus 117 dialects |
| Hailuo | No list published |
| HeyGen | Two named lists, no total |
| Higgsfield | No list published |
| invideo AI | No list published |
| Kling AI | Five, count only |
| LTX Studio | No list published |
| Pika | No list published |
| PixVerse | No list published |
| Rask AI | 135+, count only |
| Runway | No list published |
| SceneMixer | 15, plus Cantonese for dialogue |
| Synthesia | 143 named |
| VEED | 29 dub-to, 72 detectable |
| Vidnoz | 100+ and 140+, counts only |
| Vidu | No list published |
Inclusion rule. Tools with a public English site selling AI video generation or AI video dubbing. Site languages count hreflang alternates served from the host being read; x-default and alternates pointing at another host are not counted, for every vendor alike. Order. Alphabetical by tool name.
Neighbouring cells: HeyGen: dialogue language and PixVerse: dialogue language. All of them together: the cell index and the register table.
- Voice languagesNative audio on VIDEO 3.0 in five languages, with voices bound to elementscount published without a list
- Output languagesNo public list (as of 2026-09-12)no list published