Proper Subs
See Japanese word order while you watch, in English.
A free, open-source Chrome extension that transforms Japanese subtitles into literal structured English, preserving the exact Japanese word order with visible particles, so learners can map spoken Japanese directly to meaning.
- statusIn development
- licenseMIT
- built withJavaScript · Chrome Manifest V3
- Chrome extension
- Offline fallback
The LLM mode needs your own API key from a supported provider; the offline mode works without one.
What it does
Features
- 01
Structured subtitles
Live cues become literal English in Japanese order: I wa apple o every day eat. Never naturalized or reordered.
- 02
Particles inline
Original particles (wa, ga, o, ni, de, kara, to, mo, no) follow the matching English word, colour-coded and configurable.
- 03
LLM or offline
Bring your own key for Groq, Claude, OpenAI or Gemini; an offline kuromoji.js + JMdict mode works without internet or API cost.
- 04
Click-to-define
Click any word for an instant Jisho/MassDict dictionary popup.
- 05
Anki export
One-click capture of the structured text, timestamp and original Japanese.
- 06
Display controls
Font size, opacity and position sliders, particle highlighting, auto-pause and a timing offset of up to ±500 ms.
Install
Clone, then load the unpacked directory in Chrome with Developer mode enabled.
Requirements
- Chrome 110+ (Manifest V3)
- LLM API key for LLM mode (Groq, Claude, OpenAI or Gemini)
- No key needed in offline mode
git clone https://github.com/MensuraMedia/proper_subs.git
cd proper_subs- In Chrome: open chrome://extensions, turn on Developer mode, choose Load unpacked and select the cloned proper_subs folder.
- Add an API key in the extension's options page to use the LLM mode.
- Site support beyond Aniwave mirrors (Crunchyroll, Netflix, HIDIVE) is planned
Full and current instructions: the README on GitHub.
At a glance
| Browser | Chrome 110+ (Manifest V3) |
|---|---|
| LLM providers | Groq, Claude, OpenAI, Gemini (bring your own key) |
| Offline mode | kuromoji.js tokenizer + JMdict |
| Target latency | under 400 ms (Groq recommended) |
| Licence | MIT |