Captions in 99 languages, and the caveat nobody prints
Most short video is watched on mute, so captions are not decoration, they are the audio for a large share of your viewers. Katto burns them in automatically. This page also tells you where the technology stops, because that is the part you would otherwise discover after you had already committed to a tool.
Whisper large-v3 handles the transcription, so the language of your episode is almost certainly covered.
The languages with a forced-alignment model. These are the ones where a caption can highlight the exact word being spoken.
Named presets, 16 of them on the free plan. The preview in the editor is the same renderer as the export.
Installed on the render machine, not just offered in a dropdown. A font that is not on the box would silently fall back.
99 languages, 41 timed word by word
These are two different jobs and most comparisons collapse them into one number. Transcription turns speech into text, and it works across 99 languages. Alignment places each individual word on the timeline, and that needs a separate model per language. We have 41 of them. In the other 58 languages your captions are correct and readable, they simply appear a line at a time instead of lighting up word by word.
We state the split because the word-by-word highlight is precisely the effect people picture when they say animated captions. A tool advertising per-word timing across every language it supports is describing its transcription coverage and letting you assume the rest.
Where the idea itself breaks down
Japanese, Thai and Chinese are written without spaces, so there are no word boundaries to align to. Per-word timing in those languages is a fuzzy notion no matter whose model you use, and captions land at character or phrase level instead. That is not a gap in our implementation, it is the honest ceiling, and anyone claiming otherwise has not tried it in those scripts.
The preview is the renderer
Captions are burned into the video rather than attached as a track, so they survive re-uploads and platforms that ignore subtitle files. That makes the preview a promise: what you arrange in the editor is what gets baked in, down to the font. The 48 fonts in the picker are installed on the machine that renders, not merely listed, because a font that exists only in the dropdown falls back silently and you find out at export.
Fixing what the transcript got wrong
Proper nouns are where transcription fails first, and they are also the words your audience notices. A correction pass runs over names and unusual terms before you ever open the clip, and anything it missed is editable word by word in the transcript panel. Edits there feed the captions directly, so you fix a name once rather than once per clip.
Questions
Transcription and timing are two separate jobs. All 99 languages get accurate text. 41 of them also have an alignment model that places every individual word on the timeline. In the other 58, timing is per sentence, so captions appear as a line at a time rather than highlighting each word as it is said.
Because the word-by-word highlight is the effect most people mean when they say animated captions. If a tool claims per-word timing in every language it supports, it is rounding up, and you find out when you try it in your language.
Languages written without spaces do not have word boundaries to align to in the first place, so per-word timing is a fuzzy idea there regardless of the model. Captions in those languages are timed at character or phrase level, which is the honest ceiling rather than a limitation we chose.
Yes. They are rendered into the file, so they survive re-uploads and platforms that ignore subtitle tracks. That is also why the editor preview matters: what you arrange is what gets baked in.
Yes, in the transcript panel of the editor. Proper nouns are the usual culprit, and they are also corrected automatically before you ever see them, so the manual pass is usually short.
16 of the 51. The remaining 35 are on Creator. The free set is not a crippled preview, it covers the common looks including word-by-word highlight.
Try it in your own language
The free plan includes 16 of the 51 styles, which is enough to see whether the timing in your language is the word-level kind or the sentence-level kind.