# OpenSubs

> A subtitle generator that runs inside the browser tab. It transcribes a
> video's speech, translates it, styles it, and burns the result into the
> picture. The video file is never uploaded, because the product contains no
> upload endpoint.

https://opensubs.app

## How it works

- Speech recognition is OpenAI's Whisper, compiled to ONNX and run by
  transformers.js. WebGPU where the machine has it, WebAssembly otherwise.
- Three model sizes are offered, downloaded once and then cached by the
  browser: Tiny (English only, ~40 MB), Base (~80 MB), Small (~250 MB).
  Only the model is fetched from the network. The audio is not.
- Subtitles are rendered by libass — the same renderer used for the preview
  and for the burn, so what is previewed is what is produced.
- Burning in uses WebCodecs. Only the picture is re-encoded; the audio track
  is copied across untouched.
- The subtitle engine is Rust compiled to WebAssembly, shared with the
  desktop build and the command-line tool.

## What is different about it

- **There is no upload endpoint in the product.** This is stronger than a
  promise not to upload: there is no server to upload to, so a footage leak
  is not something that has to be prevented. It can be checked in the
  browser's Network tab — after the page and the model have loaded, working
  on a video produces no requests on the default on-device route.
- **No file-size limit, no queue, no watermark.** All three exist on hosted
  tools because the file has to reach a server and the server has to pay for
  the encode. Neither applies here.
- **No account for anything that runs locally.** Transcription, on-device
  translation, all 12 subtitle styles, word-by-word highlighting, trimming,
  burning into MP4 and every export format work signed out. An account is
  needed only for optional cloud translation on our own backend, priced at
  $5 for 1,000 credits, with the price of a job shown before it runs.
- **Mixed-language audio produces one subtitle file.** The spoken language
  is detected every four seconds rather than once per file, so a video that
  switches between two languages is transcribed in both. Whisper's own
  pipeline has no language detection at all; this is implemented on top of
  it.
- **Exports:** .srt, .vtt and .ass; .txt, the words alone with no timings;
  or an MP4 with the subtitles burned in.
- **Licence:** AGPL-3.0-or-later.

## Honest limitations

- Accuracy is Whisper's accuracy at the model size the machine can afford,
  and a browser is a smaller budget than a server.
- **The figures below are a floor, not field accuracy.** They are measured
  on synthesised speech, where the reference transcript is exactly the text
  that was spoken, so there is no annotator judgement in them — and also no
  accent, no overlap, no room noise and no music. Real footage scores worse.
  They are published in this form because a number with its method attached
  can be checked, and "highly accurate" cannot.

  Measured on 2026-09-09, three single-language clips, `e2e/accuracy.mjs`:

  | Model | English (WER) | Chinese (CER) | Japanese (CER) |
  |-------|---------------|---------------|----------------|
  | Base (~80 MB)  | 7.7% | 3.4% | 4.3% |
  | Small (~250 MB)| 0.0% | 1.7% | 4.3% |

  Small roughly halves the error on English and Chinese and costs about
  twice the time. On Japanese it scored identically to Base on this clip —
  a bigger model is not a guaranteed improvement per language.
- **No comparison against other tools yet.** A head-to-head against a
  competing browser-based tool is open as a tracked task and unfinished, so
  this file makes no relative claim.
- The first visit downloads a model. That is a real wait, and it is the
  price of not uploading the video.
- Cloud translation, if chosen, sends the subtitle *text* — never the video
  — to whichever provider was picked. Cloud transcription, if chosen, sends
  the audio span you trimmed to the service whose key you supplied — again,
  never the video. The on-device options send nothing and are the defaults.
- iOS 17 Safari cannot run the recogniser at all; iOS 26 works.
- **The editor changes words, not timings.** Cue text is editable; the start
  and end times are not, and cues cannot be added, deleted or merged. A file
  that is out of sync with its video needs a different tool.

## Where it runs

- In the browser at https://opensubs.app/, with nothing installed.
- As a browser extension that puts subtitles over whatever is playing in the tab:
  - Chrome Web Store: https://chromewebstore.google.com/detail/live-subtitles-in-your-br/ebiggkbolhfmhhiclmhkjkjofnflbepd
  - Firefox Add-ons: https://addons.mozilla.org/en-US/firefox/addon/opensubs/
  - Edge Add-ons: https://microsoftedge.microsoft.com/addons/detail/live-subtitles-in-your-br/ocegdknngeahbblajpajaekoekghgdnh
- As a desktop app for macOS, Windows and Linux, for long files (one video at a time).
- As a command-line tool, built from source.

Contact: hello@opensubs.app

## Pages

- [Subtitle generator](https://opensubs.app/): the tool, and what it does.
- [Burn subtitles into video](https://opensubs.app/burn-subtitles-into-video):
  hardcoding captions into the picture.
- [Subtitle styles](https://opensubs.app/styles): the twelve presets, each
  rendered live from the engine's own numbers rather than screenshotted.
- [Translate subtitles](https://opensubs.app/translate-subtitles)
- [Subtitle sites](https://opensubs.app/subtitle-sites): where to download subtitles, and making them yourself.
- [Browser extension](https://opensubs.app/browser-extension): subtitles over whatever is playing in the tab; Chrome, Edge and Firefox.
- [Transcribe video](https://opensubs.app/transcribe-video): transcribing a video to text in the browser, exported as .txt (the words alone) or as .srt, .vtt or .ass.
- [Whisper in the browser](https://opensubs.app/whisper-in-browser): how OpenSubs runs Whisper locally in the tab, and the three model sizes.
- [Subtitle editor](https://opensubs.app/subtitle-editor): correcting the text of an .srt or .vtt in the browser; timings are not editable.
- [Caption generator](https://opensubs.app/caption-generator): generating captions from a video's speech in the browser, exported as a caption file or burned into an MP4.
- [VTT to SRT](https://opensubs.app/vtt-to-srt): converting a .vtt to .srt in the browser, and what the conversion keeps and drops.
- [SRT file generator](https://opensubs.app/srt-file-generator): creating an .srt file from a video's speech in the browser, and what the file contains.
- [Bilingual subtitles](https://opensubs.app/bilingual-subtitles): putting the original and a translation on screen together, burned into an MP4 or exported, for your own video files.
- [VEED and Kapwing alternative](https://opensubs.app/veed-alternative): OpenSubs compared with VEED and Kapwing for adding subtitles (price, watermark, upload, subtitle files, translation), each fact sourced, and when either of them is the better choice.
- [Blog](https://opensubs.app/blog)
- [Privacy](https://opensubs.app/privacy): what is stored, and where.
- [Windows desktop privacy policy](https://opensubs.app/desktop-privacy): the
  separate policy for the Windows desktop app.

## Languages

Four pages exist as real documents in eight languages, translated at build
time rather than by script, each with its own URL and canonical: the home
page, Burn subtitles into video, Subtitle styles and Privacy. The home page
in each language is:

- English — https://opensubs.app/ (also the hreflang x-default)
- 简体中文 — https://opensubs.app/zh-Hans
- 繁體中文 — https://opensubs.app/zh-Hant
- 日本語 — https://opensubs.app/ja
- 한국어 — https://opensubs.app/ko
- Deutsch — https://opensubs.app/de
- Español — https://opensubs.app/es
- Português — https://opensubs.app/pt

The other three sit under the same prefix: /ja/burn-subtitles-into-video,
/ja/styles, /de/privacy, and so on. Where the tool is embedded in a page, it
follows that page, so the interface and the copy around it are the same
language.

The remaining pages — Translate subtitles, Subtitle sites, Browser
extension, Transcribe video, Whisper in the browser and the blog — are in
English. One blog post, How to add captions
to a video, also has a Portuguese version at
https://opensubs.app/pt/blog/how-to-add-captions-to-a-video.