Subtitles for your video, made in your browser.
Drop a video or a recording. The speech is recognised on this device, word by word, then cut into readable subtitles or short captions you can fix and save.
Runs in your browser. Nothing is uploaded, no signup.
Watch an example
14 seconds · a made-up example
Videos made in the studio get their captions from the voice's own word timings.Join the beta
How it works
Short and checkable: every limit links to the platform's own page, read on 28 Oct 2026.
Recognised on your device
Whisper, OpenAI's open speech model, runs inside this tab through Transformers.js. The first time, your browser downloads the model from Hugging Face (41 MB for Fast, 77 MB for Accurate) and keeps it; after that it starts at once. Your file itself is read here and never sent anywhere.
Word by word, then readable
The sound is cut into parts of at most 30 seconds, always at a quiet moment so no word is split. Each word comes back with its time. Subtitles then follow Netflix's rules — two lines of 42 characters at most, a new subtitle at the end of a sentence or after a pause — or, for short videos, one short line at a time.
Fix, check, save
Read the result through and correct any word in place: the model can mishear names, brands and numbers. Then send it to the subtitle checker for reading speed and overlaps, or save it as SRT or VTT.
What it doesn't do
- It is slower than a server: on a laptop, expect roughly the length of the video or more with the Accurate model. It depends on your computer.
- It doesn't separate speakers or translate.
- Music, noise and several people talking at once lower the accuracy.
- Your file is never uploaded; only the model is downloaded, once, from Hugging Face.
Questions
Is it really free?
Yes. The speech model (OpenAI's Whisper, open source) is downloaded once by your browser and runs on your computer. Nobody is paid per minute, so there is no limit.
Is my video uploaded?
No. The sound is decoded and recognised in your browser. Only the model itself is downloaded, from Hugging Face.
Which languages does it support?
Whisper recognises about a hundred languages; the list in the tool keeps the common ones. Accuracy is lower on fast speech, music and small languages — read and fix the text before you publish.
Sources
- 1Hugging Face · onnx-community — Whisper tiny and base with word timestamps (ONNX builds of OpenAI's Whisper)Apache-2.0; downloaded by your browser · read 28 Oct 2026
- 2Hugging Face — Transformers.js 3.8.1 — runs the model in your browserApache-2.0 · read 28 Oct 2026
- 3Netflix Partner Help — English (USA) Timed Text Style Guide: 42 characters a line, 2 lines, 20 characters a second17 a second for children; change log 19 Dec 2025 · read 28 Oct 2026
More free tools
All 25 tools- Make the videoShort video analyzerWhy does this short work?
- Make the videoScript timerHow long will this script run?
- Make the videoVideo to textNeed the words of a video as text?
- Make the videoBurn subtitlesSubtitles that stay on TikTok and Reels?
- Make the videoSubtitle checkerCan people read your subtitles in time?
- Make the videoSilence removerCut the pauses out of a talking video?
Captions on every video, timed to the word.
Vertiscale captions every video it makes on the voice's own word timings, in whole phrases — and on each version for each of your accounts.