ByteScope

Subtitle Translate

Translate a YouTube transcript or an SRT/VTT file in this tab — pick the engine, get bilingual output, upload nothing.

Your files never leave your browser — all processing is local.

Drop an SRT/VTT file, paste subtitle text, or paste a YouTube transcript you copied. Translation runs in this tab.

Drop a subtitle file

SRT, VTT, ASS, SBV, LRC — or click to browse

Translation
Compare all models

Version and platform facts

Translator and Language Detector: desktop Chrome 138+. Mobile Chrome is ✗ on every row.

APIChromeChrome AndroidEdgeFirefoxSafari
TranslatorChrome 138+ stable. Desktop only.YesNoUntestedNoNo
Language DetectorChrome 138+ stable. Desktop only.YesNoUntestedNoNo

Translator language packs are small. Gemini Nano (Prompt / Summarizer) is a much larger download and is not used here.

This browser, right now

  • Translator …
  • LanguageDetector …

About this tool

A subtitle file does not need a cloud quota. Drop SRT or VTT (or ASS / SBV / LRC), paste the text, or paste a YouTube transcript you copied yourself. A translation engine of your choice then walks the file cue by cue, on your machine: Chrome's built-in Translator API — fast, with a small language pack per pair that downloads on demand, and the default when this computer has no usable GPU — or Gemini Nano, Chrome's larger built-in model, slower per cue. With a usable GPU the default is Hy-MT2, a downloaded model that came out more accurate in our blind comparison. There is no account, no word limit from us, and the file never leaves this tab. Download translation-only or bilingual SRT/VTT when it is done.

Models

Hy-MT2 1.8B — Tencent Hunyuan-MT 2, downloaded once (about 1.3 GB) and run locally on your GPU; the default whenever a usable graphics card is present, and the engine our blind evaluation preferred over Chrome Translator for English, Chinese and Japanese directions. Chrome Translator — Chrome's built-in on-device translator with a small language pack per pair; the default without a GPU. Gemini Nano — Chrome's built-in model, slower per cue. Hy-MT2 runs on your graphics card through WebGPU: a compatible-GPU build covers older cards without shader-f16 (a GTX 10-series, say), a fast-GPU build uses f16 when the card supports it, and it has no CPU build: without a usable GPU the page switches to Chrome Translator instead. Nothing is uploaded with any of the three, and a model downloaded here is shared by every ByteScope tool.

How to copy a YouTube transcript

Open the video → expand the description → tap Show transcript → select all and copy → paste it into the YouTube tab here. Turn timestamps on before you copy; without them this page cannot rebuild cue times. This is the only legal path: the page will not fetch captions from a YouTube URL (CORS, bot detection, and YouTube's terms all forbid that). The same steps work for Translate YouTube subtitles searches — you bring the text, we restore the timestamps.

What this browser has to ship

Hy-MT2 is not a Chrome feature: it is a 1.8B model downloaded once (about 1.3 GB) and run on your graphics card, so it works in any browser with WebGPU. The other two engines are Chrome built-ins: Translator and Language Detector are Chrome 138+ on desktop, and Gemini Nano is Chrome's own on-device model, available only where Chrome provides it — unavailable is a normal answer there, not a bug. Mobile Chrome is a hard no for the built-ins. Firefox and Safari do not ship the APIs. Edge is Chromium and untested here. If this browser cannot translate, the matrix, the YouTube steps, the about, and the FAQ still render. Parse and validation run everywhere; only the translate button needs an engine. The picker never downloads anything on its own: a model or language pack that isn't on this machine yet is shown with its size first, and starts only when you say so.

Frequently asked questions

Is my subtitle file uploaded?

No. Parsing, language detection and translation all run in this tab, whichever engine you pick. The only downloads are the Hy-MT2 model the first time this machine uses it (about 1.3 GB, once), a Translator language pack the first time this machine needs that pair, or Gemini Nano itself, fetched by Chrome — and your cues are not attached to any of those requests.

Which translation engine should I choose?

With a usable GPU, Hy-MT2: it is the recommendation there, because it came out more accurate than Chrome's Translator in our blind comparison, at the cost of a large one-time download. Otherwise Chrome's Translator for almost everything: it is the fastest, and its language pack per pair is a few tens of megabytes cached by the browser. Gemini Nano is the larger built-in model, slower per cue and only available where your Chrome provides it; try it when the Translator pack for your pair produces stilted output and you have the time to spare, or when Chrome has no pack for that pair at all. All of them run on your machine and keep the file in this tab, so privacy is not the deciding factor — speed and the language pair are.

How do I translate YouTube subtitles here?

On the video: expand the description, open Show transcript, copy the text (with timestamps), and paste it into the YouTube tab. We parse the timestamps back into cues. Paste a youtube.com URL and nothing happens — that fetch is blocked on purpose.

The paste has no times / cues overlap. What now?

If there are no timestamps, YouTube probably hid them — open the transcript menu and turn timestamps on, then copy again. Overlaps and zero-length cues are reported after parse (a pasted transcript can create those). They do not block translation; fix them in Subtitle Sync afterwards if a player chokes.

Which browsers can actually translate?

Any browser with WebGPU, using Hy-MT2 — Firefox and Safari included, once the 1.3 GB model is downloaded. The built-in engines need desktop Chrome 138+ with Translator, or with Gemini Nano available; mobile Chrome, Firefox and Safari do not ship them. Edge may work as Chromium but is marked untested until someone measures it. The table stays on the page even when this browser cannot run the tool.

Can I stop a long file and continue later?

Yes. Abort stops at the next cue. Resume feeds the partial translations back in so finished cues are not sent again, with either engine. Failed batches are listed with the cue range they covered. Output can be translation-only or bilingual (translation on top, or original on top).

Related tools