Public videos that already have captions. youtube.com, youtu.be and /shorts/ links all work. You get a .txt and a .srt file, both built in your own browser.
Paste a YouTube link, take the words away as a file — TXT or SRT, no signup
Public videos that already have captions. youtube.com, youtu.be and /shorts/ links all work. You get a .txt and a .srt file, both built in your own browser.
It reads the caption track the video already has and writes it to a file — it never listens to the audio and never re-writes the words from scratch. Nothing has to play through, which is why the file is ready in seconds rather than after a recording finishes.
Two things happen when you press the button. This site asks YouTube which caption tracks the video has and hands that list to the page. Your own browser then fetches the track you picked straight from YouTube, and the file is assembled in your browser and saved by your browser. No copy of your transcript is sent to this site, no file is parked on a server waiting for you, and there is nothing to delete afterwards.
YouTube asks the browser that requests captions to pass a bot check first. Your browser does that check itself; this site never sees the result of it. One consequence worth knowing: an extension that blocks Google scripts, or a privacy browser in its strictest mode, can stop the download before a file exists. When that happens you get a message that says so.
The two buttons give you the same words in two different shapes, and the shape decides what you can do with the file.
[mm:ss]; switch to Plain text before downloading and you get the words as one continuous block with no times in it.Both files are UTF-8 text with Unix line endings. Both are named after the video — the title, cleaned up, so Me at the zoo saves as Me-at-the-zoo.txt and Me-at-the-zoo.srt.
The times are not calculated here — they come with the caption track. YouTube sends every caption line with the millisecond it starts at and how long it lasts. The SRT file copies those two numbers: the start time goes on the left, and the end time is that start plus the duration that came with the line. Milliseconds are kept, so you see 00:00:01,200 --> 00:00:03,360 rather than times rounded to whole seconds.
In the file I downloaded from Me at the zoo, the first cue reads 00:00:01,200 --> 00:00:03,360. The caption line arrived with a start of 1,200 ms and a duration of 2,160 ms; 1,200 + 2,160 is 3,360, which is the end time you see. Nothing in between was estimated.
The one place this page fills a gap: if a line arrives with no duration, the file gives it two seconds so the cue is still visible on screen. Across the three videos I measured, cues that came out at exactly 2.000 seconds: 1 of 6 on the 0:19 video, 4 of 428 on the 21:03 video, 1 of 90 on the 4:42 video. Some of those may genuinely be two-second lines in the original track — once the file is written you cannot tell the two apart, so I am not going to claim a number I cannot separate out.
The files are line-by-line, not grouped into paragraphs. YouTube stores captions as a stream of short timed lines, and that is what you get — each caption line becomes one line in the TXT and one cue in the SRT. Nothing is merged, and no sentence is split across two cues.
That is worth knowing before you open a long file: it reads like subtitles, not like an article. The 21:03 talk I measured came back as 428 separate lines. If you want flowing prose instead, switch to Plain text and download the TXT — that joins the lines into a single block with the times removed.
The SRT is consequently taller than the TXT: 428 lines of text became 1,712 lines of SRT, because every cue carries a number and a time line as well as the words.
The longest video I put through this page is a 21:03 TED talk. Its caption track came back as 428 lines, which produced a 23.0 KB TXT file (23,528 bytes) and a 34.2 KB SRT file (34,976 bytes). The SRT is about 1.5× the TXT — that ratio is the cost of the timing lines, and it held across the other videos too (0:19 gave 265 B and 415 B; 4:42 gave 4,087 B and 6,418 B).
In practice that means even a talk of this length lands as a file you can email, drop into a document or open on a phone without thinking about it. The longest single cue in it lasted 7.2 seconds and the shortest 1.0 second, which is what natural speech looks like when it is cut into subtitle lines.
Nothing here invents a caption track, so sometimes there is no file. These are the messages, quoted exactly as the page shows them:
What is not in that list: a length limit, a quota, a daily allowance or a waiting room. There is no account here, so there is nothing to run out of from your side.
Every row is one run on the live page, in a real browser, on my own connection, on 17 September 2026. The file sizes are read off the files that actually landed on disk — I clicked the button and opened what came out.
| Video | Length | Track | Lines | TXT | SRT | SRT cues | Time to result |
|---|---|---|---|---|---|---|---|
| Me at the zoo | 0:19 | English | 6 | 265 B | 415 B | 6 | 12.3 s |
| Despacito | 4:42 | English | 90 | 4,087 B | 6,418 B | 90 | 1.1 s |
| Despacito | 4:42 | Spanish | 91 | 4,107 B | 6,464 B | 91 | 1.0 s |
| Your Body Language May Shape Who You Are (TED) | 21:03 | English | 428 | 23,528 B | 34,976 B | 428 | 22.4 s |
How to read the time column: it is the whole trip from pressing the button to the transcript appearing. The two rows at about one second are videos I had already looked up recently, so the answer was still cached; the 12.3 s and 22.4 s rows were cold starts that included the full lookup.
Where my testing stops: the longest video I ran is 21:03, so that is the only place I can tell you what a long file looks like. I have not measured a length ceiling, a file size ceiling, an accuracy figure or how many languages work — I have no number for any of those, so none of them appear on this page.
Yes. Paste the link in your mobile browser and both download buttons work. On Android the file lands in your Downloads folder; on iPhone it goes to Files, and Safari will offer to open it or save it. The page is laid out for a phone screen first, so the buttons sit under the text rather than off to the side.
No. The file is built in your browser from the caption text your browser fetched from YouTube, and it is saved straight to your device. This site handles the link you paste and nothing else — see the privacy policy.
Whatever the video has. If there is more than one track, a dropdown appears above the transcript and you pick before you press the button; change it and press again to rebuild the file in that language. If there is one track, that is the one you get. I am not going to quote a count of supported languages — it is a property of the video, not of this page.
Because of the timing lines. Every cue carries a sequence number and a start/end line before the text, and the cues are separated by blank lines. On the 21:03 video that overhead took the file from 23,528 bytes to 34,976 — about 1.5×.
They match the caption track, which is what YouTube itself shows as subtitles. If the uploader's captions drift or start late, the file drifts with them, because the file is a copy of that track rather than a fresh measurement of the audio.
Not from this page — it only reads captions that already exist. If you have the video as a file, Video to Text or Audio to Text run a speech model in your browser and do not need a caption track.
No. There is no upload step, so there is nothing to keep, and no account to attach it to.
InstaScript is an independent personal project. It is not affiliated with, endorsed by, or sponsored by YouTube or Google.