SubVibe — AI subtitles for streaming video

Overlay AI-translated subtitles on streaming video, in your language — pre-translated ahead of the playhead so there's no lag, perfectly synced to playback, and cached locally so re-watching costs nothing. Or hear it translated aloud, live, in your language.

Supported: YouTube · Netflix · ZDF · Deutsche Welle · Amazon Prime Video · Udemy

Features

• Pre-translated ahead of the playhead — grabs the whole caption track up front and translates the upcoming lines before you reach them. A toolbar badge shows how far ahead it's ready.
• Perfect sync — cues are keyed to the exact playback time; scrub, pause, rewatch — they follow.
• Live Translate — hear the video in your language as it plays, in real time, powered by Google Gemini. Keeps the original speaker's own voice; 70+ languages; one click, no microphone.
• Style the original, free — apply your style and the karaoke highlight to the video's own captions, with no translation and no cost (also handy just to resync captions to the picture).
• Cached & free on replay — generated once, stored locally (IndexedDB); replays cost nothing.
• Dual subtitles — show the translation, the original line, or both stacked (great for learning).
• Karaoke highlight — words light up as they're spoken; word-exact on auto-captions, closely estimated elsewhere.
• 60+ subtitle languages (70+ for Live Translate), right-to-left support (Persian, Arabic, Hebrew, Urdu) + a bundled Persian font.
• Per-video settings — language(s), position (drag each line), size, and a sync nudge, remembered per video.
• Style presets & custom looks — Classic, YouTube, TikTok, Pill, Snapchat, Cinema and Minimal presets, plus custom font, text color, background color/opacity, outline/shadow — and a free size slider.
• Translated screenshots — right-click any page → Screenshot with SubVibe: the visible area, the full page, a dragged region or one element, with the page's own text in your language (the site's fonts and layout, right-to-left handled). Edit a translation and re-shoot, add a clean frame, then download, copy or share; shots stay on your device.
• Bring your own key (BYOK) — uses your OpenAI, Anthropic, or Google key, stored only on your device. No SubVibe servers, no accounts, no ads, no tracking.

Install

From the Chrome Web Store: 👉 Install SubVibe — then click the icon, add your API key, and pick your language.

From source:

1. chrome://extensions → enable Developer mode.
2. Load unpacked → select this folder.
3. Click the icon → paste your API key (OpenAI, Anthropic, or Google), pick your language(s).

Firefox (experimental): ./build.sh --firefox builds a Firefox package (same source, two manifest tweaks). Load it via about:debugging#/runtime/this-firefox → Load Temporary Add-on — an AMO listing is planned, which Firefox needs for permanent installs. The live audio features are Chrome-only (Firefox has no offscreen API).

On Android?

Chrome for Android can't run extensions — no extension can work there. What does work:

• Firefox for Android (planned, via the AMO listing): realistic on DRM-free sites — ZDF, DW, Udemy, and YouTube in "Request desktop site" mode. Netflix and Prime Video don't allow mobile-browser playback at all (app-only DRM), on any browser.
• Chromium browsers with extension support (Quetta, Lemur, Edge Canary) can install the normal Chrome build today — same site caveats as above.

Use

1. Play a video (turn the player's own captions on once, so SubVibe can read the track).
2. Subtitles appear over the player, pre-translating ahead — or start Live Translate from the popup's Dub tab to hear it spoken in your language.
3. Re-watch anytime — it replays from cache, free (DevTools ▸ Network shows no provider calls).

Privacy & BYOK

SubVibe has no servers of its own. Your API key (OpenAI, Anthropic, or Google) is stored locally and used to call that provider directly from your browser; only the video's caption or audio text is sent there, to translate it. Nothing else leaves your device. → Privacy policy

Open source & verifiable builds

SubVibe has no build step — it's vanilla JS/HTML/CSS, nothing minified or bundled. The files in this repo are exactly what ships and runs.

Reproduce the published package from source:

./build.sh        # → subvibe-v<version>.zip, the exact zip uploaded to the Web Store

Verify the published extension matches this source:

1. Chrome installs extensions unpacked at …/Chrome/Default/Extensions/<id>/<version>/.
2. diff -r that folder against this repo at the matching release tag (e.g. v1330.2.0).
3. A clean diff means identical code — and because nothing is minified, the diff is human-readable.

Note: the .crx Google serves is re-signed/re-packaged, so it isn't byte-identical to the upload — but the file contents are, which is exactly what the diff confirms.

Project structure

• manifest.json — MV3 config & permissions.
• background.js — service worker: IndexedDB cache + provider calls (cross-origin lives here, never in a content script).
• content/common.js — the engine: detect source, build per-language cues, render & sync the overlay.
• content/adapters/* — per-site caption acquisition (YouTube, Netflix, ZDF, DW, Prime, Udemy).
• content/subs-intercept.js — MAIN-world subtitle/segment sniffer + page-world playhead relay.
• offscreen-live.js — the Live Translate audio session (tab capture ↔ Gemini ↔ playback).
• popup.html / popup.js — settings (keys, languages, appearance, per-video).
• shared/, styles/, fonts/, icons/ — shared data, overlay styling, RTL font, icons.

License

MIT — see LICENSE.

—

Made by Nimanou.