Mirumoji¶

An open-source, self-hosted Japanese Immersion Toolkit

Drop in a video, an anime episode, a drama, or an audio clip and Mirumoji gives you clickable tokenized subtitles with instant dictionary lookups whisper-powered subtitle generation, clip saving, and Anki export. All running locally in Docker, with optional cloud GPU and LLM features

Get Started Try the Live Preview

Features¶

Interactive video player with clickable subtitles — **Interactive Video Player** → Load any local video and `.srt` subtitles, with synced, clickable Japanese lines

Word lookup dialog — **Clickable Subtitles** → Subtitles are tokenized in the server (UniDic). Click any word for readings, meanings, and contexts

Kanji breakdown in the word lookup — **Kanji Breakdown** → Every lookup breaks down its kanji with stroke counts, readings, and meaning (KANJIDIC2)

Dictionary search page — **Dictionary** → Wildcard search powered by kotobase (JMdict / JMnedict)

Dictionary example sentences with furigana — **Example Sentences** → Entries (Tatoeba) come with furigana-annotated example sentences to see words in use

Profiles & Data → Your files, transcripts, clips, and LLM templates stay organized per profile on the server

Audio transcription page — **Transcription** → Transcribe audio or generate subtitles with faster-whisper, locally or offloaded to a cloud GPU

LLM word breakdown — **LLM Breakdowns**. Optionally use Gemini / Claude / GPT API Keys (or your own local LLM server) to generate sentence breakdowns

Customizable LLM breakdown template — **Custom LLM Prompts** → Pick your provider and tailor the breakdown prompt and system message per profile.

Saved clips and Anki export — **Clips & Anki** → Save video segments with their word breakdowns and export them as an Anki deck

Desktop launcher dashboard — **Desktop Launcher** → The GUI starts and manages the docker compose application for you

How It Works¶

Services

Mirumoji Is A Docker Compose Application With Two Services

Frontend → A React app served over HTTPS that you open in any browser on your network
Server → A FastAPI backend that tokenizes Japanese, runs transcription, manages your profiles, and talks to optional cloud services

Optional Features

Two capabilities are optional and entirely under your control. Everything else (your media, profiles, clips, and database) stays in your machine

Transcription Backend → Run faster-whisper on your local NVIDIA GPU or offload it to Modal Cloud GPUs (CPU-Only Setup). See Using a GPU
LLM → Provide an OpenAI, Anthropic, Gemini, or Custom OpenAI-Compatible key to unlock AI sentence breakdowns + subtitle refinement. Without a key, those features are hidden

Get Started¶

GUI Setup

Download Desktop Launcher For Your Platform To Start, Stop, and Configure Mirumoji (Easiest)
CLI Setup

Install With pip And Run The Docker Compose Application With mirumoji up (For Those Who Prefer The Terminal)
Manual Setup

Run The Docker Compose Commands Yourself, No Launcher (More Configuration)
CLI Reference

Every mirumoji Command Explained