To generate an SRT file from a video, the speech has to be transcribed, split into short timed lines, and saved as numbered cues in a plain text file. suubbs does all three on your own machine: drop the video in, press Generate, fix the few words the model misheard, and press Export .srt. The free browser version does this for the first 60 seconds of any video; the Mac app does the whole video, however long it is.
The same five steps in the browser and in the Mac app. Only the model, the speed and the length limit differ.
The Mac app runs Whisper Large v3 Turbo through whisper.cpp on Apple's Metal, with word boundaries measured at about 3 ms. The browser version runs smaller Whisper models on WebGPU, and its word timings are looser, roughly 60 to 140 ms. For an SRT that is mostly invisible, since a cue's start and end are what a player shows. It matters more for word-by-word captions, where the highlight has to land on the syllable.
It helps to know, because most problems people have with subtitle files come from one of these four parts being slightly wrong.
An SRT (SubRip) file is plain text. Each cue is a number, a timing line, one or two lines of caption, and a blank line:
1
00:00:02,100 --> 00:00:04,400
You could type all of these by hand…
2
00:00:04,600 --> 00:00:07,200
Or you could just press Generate.
hours:minutes:seconds,milliseconds, joined by --> with a
space either side. The separator before the milliseconds is a comma. A full stop
there is WebVTT's convention, not SubRip's, and a file mixing the two is one of the most
common reasons a player shows nothing at all.Nothing in the format stops two cues from sharing time. Overlapping timestamps are how an SRT shows two subtitles at once, and the player decides how to lay them out. If the SRT you need is already inside a video file as a track, you do not have to generate one: extracting subtitles from a video shows how to copy it out, and how to tell when there is nothing to copy.
Save the file as UTF-8. Accented letters, curly quotes and anything outside English can come out as garbage in a player that guesses a different encoding, and a file that looks perfect in your editor is no evidence it will look perfect in someone else's player.
Every SRT generator is one of these three. They differ in where the work happens and who ends up holding a copy of the video.
| Typed by hand | Upload to a web service | Transcribed locally | |
|---|---|---|---|
| Where the video goes | Nowhere | To the service's servers | Nowhere |
| Your effort | Every word and every timestamp | Upload, wait, correct | Correct the words it misheard |
| Timing | As careful as you are | Depends on the service | Per word, about 3 ms in the Mac app |
| Best for | A few lines, or a short clip | When uploading the footage is fine | Anything you would rather not upload |
By hand is perfectly possible for a short clip. On a Mac, open TextEdit,
switch the document to plain text with Format ▸ Make Plain Text, type the cues in
the shape above, and save with a .srt extension. Then check in Finder that it
did not quietly become captions.srt.txt. The slow part is the timestamps:
for each line you scrub the video, note when speech starts and stops, and type both down to
the millisecond.
Uploading to a web service is how most online subtitle generators work: the video goes to their servers, their speech model transcribes it, and you download the result. That is fine for plenty of footage. It is not fine for client work under NDA, unreleased material or interviews you promised to keep private, which is the case an offline subtitle generator is built for.
Transcribing locally is what suubbs does. The speech model runs on your own hardware, so there is no upload, no queue and no per-minute charge. You can also run whisper.cpp from Terminal yourself, which is free and produces an SRT; what you give up is the editor afterwards, which Whisper on a Mac, without the terminal lays out line by line.
The format itself has no style section. What it has is a handful of tags that many players read, and nothing beyond them.
SubRip has nowhere to declare a style for the whole file. What many players do honour
is a small set of HTML-like tags inside a cue: <b>, <i>
and <font> with a colour, a typeface and a size. When you have styled a
document in suubbs, the SRT export writes those tags for whatever differs from the default
look, and only that, so an unstyled document still exports as a bare file every tool
expects. A player that does not parse them shows them as literal text, which is the risk
of using them at all. The size tag is written in pixels against the video's height —
subtitle font size has the conversions.
What SRT cannot carry, with or without tags: the backing box, the outline or shadow, the
position on screen, letter spacing, fades and word-by-word highlighting. suubbs names
whichever of those you used when you export, rather than letting you find out on the
upload. If they matter, export Advanced SubStation (.ass), which keeps them (a
backing box currently needs a one-line override at burn-in) and is what ffmpeg uses to burn subtitles into
the video, or WebVTT for a web player, which holds most of them as CSS — WebVTT styling shows what it writes. Where a caption sits is
covered in subtitle position, including the
{\an8} tag some players read in an SRT.
SubRip, unstyled exists for the opposite case. DaVinci Resolve treats a cue that arrives carrying its own tags as hand-styled, so the subtitle track's style stops reaching that one line while every other line takes it. A file that states nothing lets the editor style all of it at once. Use it whenever the styling will be done in the editor, as in subtitles for DaVinci Resolve and Final Cut Pro.
suubbs writes five files from the same timings, so choosing one never means transcribing again.
<track> element.
Already have an SRT? Converting SRT to VTT is two small
changes.Check these in order. The first three cover almost every file that "does nothing".
captions.srt.txt is a text file, not a
subtitle file. Finder hides extensions by default, so check with Get Info.Short answers, including where the format simply gives up.
Transcribe the speech, split it into short timed lines and save them as numbered cues. In suubbs that is: drop the video in, press Generate, correct the words it misheard, and press Export .srt. Transcription runs on your own machine, so the video is never uploaded.
Yes. The browser version of suubbs needs no account and transcribes the first 60 seconds of any video, with every editing and export control working on that minute. The $9.99 Mac app transcribes the whole video. Typing an SRT by hand in a plain-text editor is also free, just slow.
A plain text file listing numbered cues, each with a start and end timestamp and one or two lines of text. It is the oldest and most widely supported subtitle format there is — every video player, every editor and every social platform reads it.
Colour, typeface, size, bold and italic, partly: many players read a few HTML-like tags inside a cue, and suubbs writes them when you have styled the document. Position, outlines, boxes, fades and word highlighting have no place in the format at all. If those need to survive, export SubStation Alpha (.ass) instead.
WebVTT is the web-native descendant of SubRip. The syntax is nearly identical — VTT uses a dot before milliseconds where SRT uses a comma, and adds a WEBVTT header. Use VTT for HTML video track elements and SRT for everything else.
That is called burn-in, and it re-encodes the picture. Export a styled .ass file and run ffmpeg -i input.mp4 -vf "subtitles=captions.ass" -c:a copy output.mp4. suubbs does not have a one-click burn-in button yet. The full walkthrough is in how to burn subtitles into a video.
Sixteen: English, Spanish, French, German, Italian, Portuguese, Dutch, Polish, Russian, Japanese, Korean, Chinese, Hindi, Arabic, Turkish and Ukrainian. The Mac app can detect the language for you, and it is worth letting it if you are not sure: telling the model the wrong language produces a translation rather than an error.
That is the whole point of suubbs. Transcription runs on your own machine in both the free browser version and the Mac app, so the video never leaves it.