How-to
Getting Subtitles Out of YouTube
Subtitles are text files stored separately from the video, and most of what YouTube offers is automatic captions — machine transcription, not something a human wrote. They are genuinely useful and they are not a script. We do not serve them; the reliable route is a command-line tool, and the command is below.
Two kinds of subtitles, and the difference matters
| Uploaded captions | Automatic captions | |
|---|---|---|
| Made by | The creator or a service | Speech recognition |
| Accuracy | Usually high | Variable, worse with accents and noise |
| Punctuation | Present | Sparse or absent |
| Speaker labels | Sometimes | No |
| Availability | A minority of videos | Almost everything with speech |
Automatic captions also include machine translations into a very long list of languages — we counted well over a hundred on a single test video, all generated from the English recognition rather than from the audio.
That stacking matters. A translated automatic caption is a machine translation of a machine transcription, and errors from the first stage arrive intact in the second.
The command that works
Using yt-dlp, which is free and available on every platform. This fetches English subtitles as SRT without downloading the video:
yt-dlp --skip-download --write-auto-subs --sub-lang en --convert-subs srt <url>
--skip-download— text only, no video.--write-auto-subs— include automatic captions. Use--write-subsfor creator-uploaded ones only.--sub-lang en— the language.--list-subsshows what exists.--convert-subs srt— SRT rather than the platform's native format.
We ran exactly this while writing and got a 1.2 KB SRT file back, correctly timed. On a video that is mostly music the content is largely [Music] markers, which is an honest reflection of what the recognition heard.
SRT, VTT and which to ask for
Both are plain text with timings, and the choice is about where the file is going.
SRT is the older and more widely supported format. Media players, editors and most software accept it, and a file named alongside a video is picked up automatically by many players.
VTT is the web format, used by browsers for captions on a page. It supports styling and positioning that SRT does not, which nothing outside a browser cares about.
Ask for SRT unless you are putting captions on a web page. Converting between them afterwards is trivial, and starting with the right one saves the step.
Both are readable text you can open in any editor, which is worth knowing: fixing a recognition error in an automatic caption takes a moment and no special tool.
What subtitles are genuinely good for
- Searching a long video. A transcript makes a three-hour talk searchable in a way the video is not.
- Quoting accurately. Timings let you cite a moment rather than paraphrasing.
- Study and language practice. Reading along with audio is a well-established method, and imperfect text still works for it.
- Accessibility. The reason captions exist, and the use with the strongest protection in law almost everywhere.
- Drafting a summary. Rough text beats re-watching to find the part you half-remember.
What they are not is a publication-ready transcript. Automatic captions need editing before anything is quoted verbatim, and the errors are frequently plausible rather than obvious — a wrong word that reads perfectly well is harder to catch than gibberish.
Limits worth knowing before you start
Four things people expect and do not get.
Not every video has them. Automatic captions need recognisable speech; music, heavy noise and some languages produce nothing.
Speaker identification is absent. A conversation arrives as one undifferentiated stream of text, which makes interviews harder to use than they look.
Timings are per caption, not per word. Fine for navigation, not precise enough for tight editing without adjustment.
Other platforms mostly do not offer this. Subtitle availability outside YouTube is thin, and where it exists it is usually creator-uploaded rather than automatic.
Rights, briefly
Subtitles are part of the work, and the same analysis applies to them as to the video.
Using a transcript to study, search or quote is the ordinary case and well supported — the four factors are here. Republishing someone's transcript as an article, or feeding a catalogue of them into something you sell, is a different question.
Creator-uploaded captions represent real work — often paid for — and treating them as free text because they are technically accessible is the mistake to avoid. Automatic captions are machine output from a copyrighted recording, which does not make the underlying work any less protected.
For anything published, the safe habit is the same as with clips: use what you need, attribute it, and ask when the stakes are real.
Why we do not offer subtitles, having tested that we could
The tooling we already run handles subtitles perfectly well. We verified it while writing this: one command, no video downloaded, a correctly timed SRT file back in about a second. Adding a button would not be difficult.
We have not, for reasons that are about the shape of the feature rather than the engineering.
The first is the choice explosion. A single video offered automatic captions in well over a hundred languages, plus any the creator uploaded, in several file formats each. A useful interface for that is a language picker and a format picker — two more decisions on a page whose entire point is paste, glance, tap. The ones who want subtitles would be served; everyone else would meet a more complicated product.
The second is that the audience for subtitles is largely the audience best served by a command-line tool anyway. People extracting transcripts usually want them in bulk, want specific languages, want to script the process — all things a single-file web page does badly and a proper tool does well. Publishing the command is more useful to them than a button would be.
The third is expectation. ‘Download subtitles’ sounds like getting a transcript, and what mostly exists is machine recognition with no punctuation and no speaker labels. A button implies a finished thing; an article can explain what the file actually contains before someone builds a workflow on it.
The honest summary is that this is a feature we decided against rather than one we lack. If enough people ask, the calculation changes — but the version we would ship would be one language and one format with the limitations stated, not the picker-heavy interface the full capability implies.
There is a wider principle behind this that shows up elsewhere on the site. Every feature we add is a decision someone has to make before getting their file, and the cost of a choice is paid by everyone who did not need it. That is why there is no frame rate picker, no crop, no trim and no subtitle language list — each would serve a real minority and slow down the majority, and pointing the minority at a better tool costs them less than the alternative costs everyone else.
Frequently asked questions
How do I download YouTube subtitles as SRT?
Use yt-dlp with --skip-download --write-auto-subs --sub-lang en --convert-subs srt. The full command is in this article.
What is the difference between automatic and uploaded captions?
Uploaded captions are written by the creator and usually accurate. Automatic ones are speech recognition — no punctuation, no speaker labels, variable accuracy.
Should I take SRT or VTT?
SRT unless you are captioning a web page. It is older, more widely supported, and picked up automatically by many players.
Are the translated captions any good?
They are machine translations of a machine transcription, so errors from the first stage carry into the second. Useful for gist, not for quoting.
Do other platforms offer subtitles?
Mostly not. Availability outside YouTube is thin, and where it exists it is usually creator-uploaded rather than automatic.
Do you offer subtitle downloads?
No. The capability exists in the tooling we use, but a useful interface would need language and format pickers, and the people who want transcripts are better served by a command-line tool.
VidKeep runs in your browser — paste a link, pick a quality, keep the file. No account, no app.
Open VidKeepLast updated: 2026-08-01. We revise our guides as the platforms change.
