A translated website is not a translated product
Kling AI declares 20 same-domain site languages and publishes no output-language list. SceneMixer declares 14 site alternates and publishes lists for both output and dialogue. The widest site coverage and the widest product coverage are not the same vendor. As of 2026-09-12.
| Tool | What is published | Detail |
|---|---|---|
| Kling AI | 20 site languages | No output list published |
| SceneMixer | 14 site alternates | 15 output, 15 voice plus Cantonese |
| Six other tools | None declared | Nothing published in any column |
Inclusion rule. Site languages are counted from hreflang alternates served on the vendor's own domain. Alternates pointing elsewhere and x-default are not counted, for every vendor alike. Order. By declared site languages, most first.
1A marketing site and a generation pipeline are translated by different people
Adding a language to a website is a content job: copy, a menu entry, an hreflang tag. Adding one to a generation pipeline means the model has to write in it, speak it, and keep it consistent across a series. Those tasks have almost nothing in common, and no reason to arrive at the same number.
So a reader who takes a language switcher as evidence that output will come back in that language is reading the wrong artefact. The switcher is evidence about where a vendor is selling, which is useful and is a different fact.
2The gap runs in both directions
One tool here has more site languages than anything it publishes about its product. Another has fewer site alternates than the languages it says its output supports. Neither ordering is surprising once the two jobs are separated, and either ordering looks like an inconsistency if they are not.
This register keeps three columns for exactly that reason. Collapsing them into a single language count would produce a number that is wrong for every reader in a different way.
3What a declared alternate actually promises
It promises that a URL exists and is meant for speakers of that language. It does not promise that the page is fully translated, that support is available in it, or that anything the product produces will be in it.
The one spot check in this register found a price page genuinely served in Japanese rather than a machine-swapped English page. That is one page on one vendor on one day, recorded as exactly that.
4The three columns behind this question
| Column | How many of the two tools publish a list |
|---|---|
| Dialogue language | 1 of 2 tools |
| Document language | 1 of 2 tools |
| Site language | 2 of 2 tools |
Inclusion rule. The three columns this register keeps. A column stays listed even when neither tool names a list, because silence in a column is itself recorded. Order. Alphabetical by column name.
- Site languages20 languages: ar, de, en, es, fil, fr, hi, id, it, ja, ko, ms, nl, pl, pt, ru, th, tr, uk, videclared as same-domain alternates
- Site languages14 languages: ar, de, en, es, fr, id, it, ja, ko, nl, pt, ru, tr, ukdeclared as same-domain alternates
- Output languagesNo public list (as of 2026-09-12)no list published
5Sources
Counts and lists are read from the vendor pages named in the register, checked 2026-09-12. The counting rule is on how counted. Nearby: Counts vs lists, The silent six.