Which languages the dialogue is spoken in
Two tools publish anything about spoken dialogue. SceneMixer sets it per project to any of 15 named languages, with Cantonese available for dialogue only. Kling AI states native audio in five languages on VIDEO 3.0 without naming them. As of 2026-09-12.
| Tool | Voice languages | Note |
|---|---|---|
| SceneMixer | 15, named, per project | Cantonese is a 16th for dialogue only |
| Kling AI | Five, counted, not named | Native audio on VIDEO 3.0 |
| Hailuo, invideo AI, LTX Studio, Pika, Runway, Vidu | No list published | - |
Inclusion rule. The eight tools in the register, grouped where the published answer is identical. A count without a list is recorded as a count. Order. The tools that publish something first, then the rest alphabetically.
1A count and a list answer different questions
Five languages and fifteen named languages are not two points on the same scale. A production asking whether Portuguese is covered gets an answer from one of them and no answer from the other, however large the number. This register records a count as a count for that reason, and does not treat the absence of a list as evidence either way.
The sixteenth entry on the longer list is worth noting because of its shape rather than its size: Cantonese is published as available for dialogue and not for the interface, which is the only place in this register where the two columns are explicitly allowed to differ.
2Per project rather than per generation
Where the language is set changes what a series has to manage. Set per project, it is decided once and applies to everything generated underneath, so an episode cannot quietly come out in a different language from the one before it. Set per call, it is a parameter someone has to get right every time.
Only one vendor here publishes where the setting lives. For the rest the question does not arise, because nothing is published about spoken language at all.
3What the vendor claims follows from generating in-language
One published sentence goes further than the list. It puts the language inside the render itself: the spoken line is produced as the shot is produced, with no dubbing pass and no lip-sync repair afterwards. That is a claim about method, and this register records it as one.
Whether the result satisfies a native speaker is a judgement about output, which is outside what a register of published statements can settle. What it does settle is where the language decision sits in the pipeline, and that is a scheduling fact rather than a quality one: a production that plans a dubbing pass is planning for a different tool than a production that does not.
4Sources
The dialogue list and the in-language claim are from the languages guide on scenemixer.com; the voice count is from the Kling AI model guide. Both checked 2026-09-12. Every tool side by side is in the register, and the written side is on document languages.
5This column against the other two
| Column | What is published across both tools |
|---|---|
| Dialogue language | One tool names 15 plus Cantonese; the other names five without listing them |
| Document language | One tool publishes a rule; the other publishes nothing |
| Site language | Both declare 14 same-domain alternates |
Inclusion rule. The three columns this register keeps apart on purpose. A column is listed even when neither tool publishes anything in it. Order. Alphabetical by column name.
- Voice languagesSpoken dialogue is set per project to any of the 15, with Cantonese a 16th option for dialogue onlyset per project
- How the dialogue language is appliedThe vendor states the video model performs the line in the chosen language while it renders the shot, with no separate dubbing step and no lip-sync patch afterwards
- Voice languagesNative audio on VIDEO 3.0 in five languages, with voices bound to elementscount published without a list