01 / WHAT THE WORDS ON SCREEN ARE
The words on screen are a caption track.
When a TikTok video has words running across it, those words usually exist as a caption track: a text file attached to the video, either written by the creator or generated by TikTok. That track is the first thing the TikTok transcript tool looks for, and when it finds one, the text you get back is the text stored in it.
The distinction matters more than it sounds. A caption track has already been turned into text by someone else, so a transcript read from it inherits that wording and those timings. Nothing is written from scratch, and nothing is guessed from the sound.
But plenty of TikTok videos have no track at all, and the page does not treat that as a dead end. In the same request it downloads the video, keeps the audio, and has the words written down by speech recognition. Two very different routes, one result box. That is why the result says which route ran, and why it is worth knowing which one you are on before you judge the text.
02 / THE LINK
The link is the whole input.
In the app, Share hands you the Copy link option; in a browser, take the address from the bar. Paste that into the box at the top of the TikTok page, then select Get the text. There is no file to prepare and nothing to install.
The box takes full video links, and it takes the short forms the app hands out: vt.tiktok.com, vm.tiktok.com and the /t/ share links. Those short links are followed to the video for you, so the copy you get from a phone is fine as it is. The numeric video ID on its own is accepted too, which helps when the number reached you without a full address.
Two things are not videos, and the page says so rather than inventing filler: a photo post, and a profile address. That answer is more useful than a silent failure, because it separates a link that cannot work from a link that failed for a moment. The link forms and the result's source line are covered in more detail in the walkthrough on how to get a transcript of a TikTok video.
03 / A VIDEO WITH A CAPTION TRACK
A video with a caption track is the quick case.
When the track exists, the words are already settled. The page reads the track and returns its lines in order, with the timings that came attached to them. Nothing has to be listened to on this route, and nothing has to be written down from the audio.
It is also the route without a length limit. A long video that carries a caption track is not a problem here: the restriction people usually run into on a twenty-minute video applies to the audio route, not to this one.
04 / A VIDEO WITHOUT ONE
A video without a caption track is written from the audio.
No track means speech recognition, and speech recognition means the audio is read and the spoken words are written out. You still get a transcript; the difference is where the words come from, and how long the page needs to work through them. Longer videos take longer on this route.
Two limits belong here rather than in a footnote. Speech recognition handles up to 20 minutes per video, and a clip that runs longer comes back with a message instead of half a transcript. The other limit is the material itself: a clip that is mostly music is a poor fit, because the model may return a repeating line or a stock phrase rather than real words. Clips that are mostly talking work best.
If a video has no captions and the audio is music from end to end, no transcript tool has much to work with. Knowing that before you paste the link saves a confusing result.
05 / WHAT COMES BACK
What comes back, and which file to take.
The result names its source: Caption track, or Speech recognition. Keep that line in mind, because the two can disagree. One is the caption track that came with the video; the other is what the audio sounds like to a model, so a name or a number may be spelled differently between them.
From there, four ways to take the words away. Copy puts the text on the clipboard and creates no file. TXT is the readable script with no clock in it. SRT is the subtitle file an editor expects: numbered cues, each with a start and an end time. VTT holds the same cues in WebVTT form, for a web player. When a result arrives without timings, SRT and VTT stay switched off and the text can still be copied.
None of it needs an account or an extension.
When the answer is a sentence, not text
Four messages are worth recognising. A link that is not a video link comes back saying it does not look like a TikTok video link. A video that will not load right now comes back asking you to try again or try another public video. A video with no track, whose audio has no recognisable speech, says exactly that. And a video with no track whose audio runs past the twenty-minute ceiling says the audio is longer than the page transcribes in one go.
06 / QUESTIONS
Can I turn the on-screen subtitles into a text file?
Yes. When the video has a caption track, that track is read. When it has none, the audio is transcribed instead and the result says so.
Does the page need the video to have captions?
No. A caption track is read as it stands; without one, the words are written from the audio.
Is there a length limit?
A caption track has no length limit here. Speech recognition handles up to 20 minutes per video.
Which file keeps the timing of each line?
SRT and VTT. TXT is the reading copy and carries no clock at all.
READY WHEN YOU ARE
Paste the link you copied.
Caption track or no caption track, the page returns the words and the file you can keep.
Get the text from a TikTok link ↗