TikTok transcript
in about 20 seconds
Paste a public TikTok link and get back what was actually said, as text you can copy, save as .txt, or turn into .srt subtitles. No account, no app, nothing to install.
Timed, not guessed
An 8-second clip came back in 9.1 seconds; a 4 min 44 s clip in 24.2 seconds. Both timed here on 9 November 2026. We publish the stopwatch, most tools just say "instant".
SRT and VTT included
Real timestamps, exported as subtitle files that drop straight into CapCut, Premiere or DaVinci. No paywall on this one.
No account, ever
No sign-up, no app, no daily cap. Each visitor gets five transcripts an hour so the server stays usable for everyone else.
What a TikTok transcript actually is
A TikTok transcript is the spoken part of a video written down as text. It is not the caption the creator typed under the post, and it is not the hashtags. It is only what was said out loud, in the order it was said.
There are two places people look for it, and they are easy to confuse. The first is TikTok's own auto-captions, which appear on some videos when the creator turned them on. Those are useful on screen but you cannot download them as a file, and half the videos do not have them. The second is a transcription tool that listens to the audio and rebuilds the text from scratch, which is what this page does.
Three ways to get one, compared honestly
| Method | Time for a 60s clip | Export as file | Works on every video |
|---|---|---|---|
| Type it by hand while listening | 8 to 15 minutes | Whatever you make | Yes |
| TikTok's built-in captions | Instant, if they exist | No | No, only when the creator enabled them |
| This tool | 9 to 24 seconds (measured 9 Nov 2026) | .txt, .srt, .vtt | Yes, as long as there is speech |
What we measured
Three runs made from a desktop browser (Chromium 131, 1440×900) on 9 November 2026, against the machine that serves this page, one transcription at a time. Audio is capped at the first 90 seconds, so a long video costs about as much as a 90-second one.
| Source video | Length | Text back | Time to the card |
|---|---|---|---|
| English talking-head, @jacobarthurr900 | 4 min 44 s | 1,764 characters (first 90 s) | 24.2 s |
| English 8-second clip, @mrbeast | 8 s | 107 characters | 9.1 s |
| Music-only post, @tiktok | short clip | "no speech detected" | 4.1 s |
An earlier round of rows was dropped from this table. The post behind one of them no longer resolves at all today, and the others were never recorded with a link, so nobody can re-check those numbers, and a number nobody can re-check is not evidence. Times move with server load and clip length. The cap, the five-an-hour limit and the stopwatch do not.
What it will not do
This section is longer than the sales pitch on most tools, on purpose. Knowing the edges saves you from blaming the tool for something it never promised.
- Videos over 90 seconds are cut at 90 seconds. Only the first stretch of audio is processed. A ten-minute podcast gives you its opening minute and a half, nothing more. The card does not label the cut: it appends the post’s full length after the character count (a 4 min 44 s post showed “· 284s”) and the subtitle file simply stops at 00:01:30,000. If those two numbers disagree, the shorter one is the transcript.
- A video with no speech returns nothing. Music-only clips, dance trends and silent montages come back as "no speech detected". That is the tool working, not failing.
- Text burned into the picture is not in the transcript. If the creator typed words onto the video, those pixels are never read. Only audio is.
- It guesses the language, and it can guess wrong. The model reads the language off the audio and prints its guess on the card (EN, JA, RU…). Two short English clips, one on 4 November and one on 9 November 2026, both came back as RU and the words came back as Cyrillic-looking nonsense instead of English; the 4 min 44 s English clip came back as EN and readable. No translation is offered, so a wrong guess has to be caught by you.
- Every failure uses the same sentence. “Could not read this video. Try again in a minute.” covers a dead or region-locked post, a link the resolver cannot parse, an audio track that will not extract and a model error. Only “no speech detected” is specific. A retry does not fix a post that no longer exists.
- Five transcripts per visitor per hour. The model needs about 500 MB of RAM while it runs and the server does one at a time, so this keeps the queue short instead of letting everyone wait. The full list of limits is on the limits page.
- Accuracy drops with noise. Heavy background music, strong accents and fast speech all cost words. The result is readable, not court-ready.
Where the output goes
Every result gives you three choices, and they are different formats for different jobs, not three copies of the same thing.
- Copy text puts the whole transcript on your clipboard, one paragraph. Right for pasting into a doc, a chat or an AI tool.
- .txt is the same text as a plain file, for archiving or handing to someone else.
- .srt and .vtt are subtitle files with a timestamp on every line. Import them into an editor and the subtitles line up with the video without retyping a single timecode.
If all you wanted was a quote, the first button is enough. If you are rebuilding the video with your own footage, the subtitle files are the ones that save the hour.
All three files are put together in your browser from the timestamps the server already sent, so there is no second request and no download queue. Checked on 9 November 2026: the 8-second clip gave a 193-byte .txt, a 358-byte .srt and a 366-byte .vtt; the 4 min 44 s clip gave a 2,979-byte .srt with 36 cues, the last one ending at 00:01:30,000. The .srt writes a comma before the milliseconds and a blank line between cues; the .vtt opens with WEBVTT and uses a full stop, which is what editors expect.
Frequently asked questions
The questions people actually send us, including the awkward ones.
How do I get the transcript of a TikTok video?
Open the video in the TikTok app, tap Share, then Copy link. Come back to this page, paste the link into the box and press Extract transcript. The text appears in the result card, usually within twenty seconds. There is nothing to install and no account to create.
Can I download the transcript as subtitles?
Yes. Every finished transcript carries two extra buttons: .srt and .vtt. Both keep the original timestamps, so the text stays in sync with the video when you import it into CapCut, Premiere Pro, DaVinci Resolve or any player that reads subtitle files. The plain .txt button is there for when you only want the words.
Why does it stop at 90 seconds?
Because transcribing audio is the expensive part. A two-core machine spends 19 to 24 seconds on 90 seconds of audio, and letting a ten-minute video through would hold the only processor while everyone else waits. The 90-second cap keeps the queue moving. Longer videos still work, you just get the opening stretch, and the card prints the full length of the post so you can see how much was left out.
It says no speech was detected. What happened?
The video probably has no talking in it. Music-only clips, dance videos, silent montages and text-on-screen videos all return that message, and it is accurate rather than a bug. A second cause is a video whose audio track failed to download, which usually clears up if you retry once.
Is there a limit? Do I need an account?
No account, no email, no card. The only limit is five transcripts per visitor per hour, counted per IP address. It exists because the speech model takes several hundred megabytes of RAM while it runs and the server handles one job at a time, so a queue is better than a crash.
Does it keep the original language?
It tries, and it prints its guess on the card. The language is read off the audio, so the text comes back in whatever language the model picked, and that is not always the language being spoken. Two 8-second English clips, on 4 and 9 November 2026, both came back as RU with Cyrillic-looking nonsense; a 4 min 44 s English clip that same day came back as EN and readable. Short clips, music and heavy noise are where it goes wrong most. Translations are not offered, so if the guess is wrong you get the wrong language, not a translated version of the right one.
The card says it could not read the video. Which part failed?
The card cannot tell you, and that is a weakness rather than an answer. “Could not read this video” is shown for a deleted or region-locked post, for a link the resolver cannot parse, for an audio track that will not extract and for a model error. Only “no speech detected” is a specific message. First open the link in a normal browser and check the video still plays; if it plays and the card still fails, the post is outside what the resolver can read and a retry will not help. More failed attempts are covered on the download not working page.
Will the words burned into the video show up?
No. On-screen text is part of the picture, and this tool only listens to sound. If a creator typed their whole script onto the video and never spoke it, the transcript will be empty even though the screen is full of words.
Is this allowed? What about copyright?
Reading what someone said is not the same as republishing their video, but the words still belong to whoever wrote them. Quote with credit, do not pass someone else's script off as your own, and check the rules of wherever you plan to post. ClipDocky is not affiliated with TikTok.