Open the studio in a browser tab, drop in a video or an audio file, and press Generate: speech recognition runs on your own computer and writes word-timed subtitles you can edit, style and export as SRT, VTT, ASS or word-level JSON. There is no account, no watermark and no upload. The one limit is length — the free browser version captions the first 60 seconds of whatever you give it. The whole video is what the $9.99 Mac app is for.
The free version is not a cut-down editor. It is the same studio with a time limit, and every row below is taken from the code that runs it.
| Free, in the browser | Mac app, $9.99 once | |
|---|---|---|
| How much gets captioned | The first 60 seconds of any file | The whole file, however long |
| Account or sign-up | None | None — the only thing kept is your email from the checkout |
| Watermark | None | None |
| Where your video goes | Nowhere — read from disk into the tab | Nowhere — read on your Mac |
| Speech engine | Whisper via Transformers.js, on WebGPU or WebAssembly | Whisper via whisper.cpp, on the Mac's GPU |
| Word timing | About 60–140 ms out | About 3 ms, measured against known speech onsets |
| Spelling hint for names | No | Yes — a "Names and jargon" field |
| Spoken language | You choose it from a list | You choose it, or let it detect |
| Editing, styling, retiming | All of it | All of it |
| Exports | SRT, unstyled SRT, VTT, ASS, word-level JSON | The same five |
The limit is deliberately time and nothing else. A watermark would be pointless — the thing you take away is a text file, and a watermark line in a text file is a five-second deletion — and locking features would mean the free version no longer showed you the product. So the trial does everything on its minute exactly as the paid app does, and simply does not transcribe past it.
From an empty tab to a subtitle file in your downloads folder.
One detail at the boundary: a cue that starts inside the first minute and runs past it is kept whole, rather than having its sentence cut off at 60.0 seconds. A cue that starts after the minute — say, one you typed in by hand — is left out of the export.
"Free" covers very different deals. These are the questions that separate them, with suubbs' own answers.
Many web subtitle tools transcribe on a server, which means a copy of your file sits on someone else's machine under someone else's retention policy. Here the speech model runs inside the tab, so there is no server to send the video to. That matters most for footage that is not yours to share: client work, unreleased material, interviews.
Free tools ration something — minutes, exports, resolution or features. It is worth knowing which before you spend an hour editing. Here it is length: 60 seconds, and nothing else held back.
A tool that only hands back a re-encoded video with the captions drawn in gives you nothing to upload to YouTube, nothing to correct later and nothing for an editor to restyle. A subtitle file keeps those options open. This one exports five kinds, and can still produce burned-in captions through the ASS file.
Accuracy gets the attention, but loose timing is what makes captions feel wrong — a word highlighted after it was said. The browser build's word timings land within about 60–140 ms; the Mac app, which snaps each word onto the waveform, measures about 3 ms. For plain subtitles the browser's timing is fine. For word-by-word captions the difference is visible.
The browser studio offers sixteen spoken languages and asks you to pick one. Pick the wrong one and Whisper tends to hand back a translation rather than an error, so set it before you press Generate. Proper nouns are where every speech model slips; the Mac app takes a list of names and jargon up front, and in either version a wrong word is a click to fix.
Privacy claims are cheap. This is one you can verify yourself in about fifteen seconds.
Open the studio, then open your browser's developer tools and switch to the Network tab. Drop in a video and press Generate.
You will see the model weights come down from the Hugging Face CDN the first time, and the site's own cookieless page-view analytics. What you will not see is any request carrying your video, its audio or its transcript — no upload, no chunked POST. A browser tab cannot quietly send data behind your back; every request it makes is listed right there.
The Mac app keeps the same guarantee — the model downloads once, and then transcription runs on the machine with the wifi off — but a native app is harder to inspect than a web page. That is part of why the free version exists: it is the easiest way to test the claim the paid one makes.
Four Whisper models, from a quick 42 MB download to an 800 MB one.
| Model | Download | Use it when |
|---|---|---|
| Tiny (English) | ~42 MB | Clean, single-speaker English and you want speed |
| Base (multilingual) — default | ~78 MB | A quick first look; expect slips on names and crosstalk |
| Small (multilingual) | ~250 MB | Base is missing words — the best accuracy per megabyte |
| Large v3 Turbo (multilingual) | ~800 MB | You want the best a browser can do, on a machine with real GPU memory |
Two more settings are worth knowing. Decoding can switch from greedy to beam search, which is several times slower and helps most on unclear speech. Translate to English asks Whisper to write English subtitles straight from speech in another language. Since the trial is a minute long, it is a good place to try a larger model and beam search on your hardest footage and see whether the extra wait pays off.
If the free minute is all you need — a short social clip, a trailer, a quick test of accuracy on your own voice — you are done. If you need the rest of the video, the Mac app runs Whisper Large v3 Turbo through whisper.cpp, with Silero voice detection and waveform-snapped word timing, and it captions the whole file. The Whisper on Mac guide explains that pipeline, and the offline subtitle generator page covers what "nothing uploaded" means there.
Including the ones where the answer is a limitation.
Yes. The browser studio costs nothing and needs no account. It is a trial in one specific way: it captions the first 60 seconds of a file. Styling, editing, retiming and every export work on that minute exactly as they do in the paid app. Captioning whole videos is what the $9.99 Mac app is for.
Any length will open, but the browser version transcribes only the first 60 seconds. A cue that starts inside that minute and runs past it is kept whole. For anything longer, the Mac app captions the whole file.
No. There is no sign-up, no email and no login. Open the studio and start.
No. suubbs gives you a subtitle file — .srt, .vtt, .ass or word-level .json — not a re-encoded video, so your picture is never touched and nothing is stamped into it.
Chrome and Edge are the ones to use, because they support WebGPU and run the model on your graphics card. Safari and Firefox work but fall back to WebAssembly, which is roughly five to ten times slower.
Yes. The studio opens audio as well as video, so a podcast or voice recording works the same way — in the browser, the first 60 seconds of it. See the audio to SRT guide for the details.
One download of the speech model the first time — about 78 MB for the default Base model, up to about 800 MB for Large v3 Turbo. The browser caches it, so later runs do not download it again. Your video is never part of that traffic.
Every speech model mishears names, jargon and crosstalk. Try the Small model or beam search on the same minute, make sure the spoken language is set correctly, and fix what is left by editing the line. The Mac app also takes a list of names and jargon before it starts.
[nothing uploaded]