Saitenka¶

Learn Japanese from the video you're already watching.

Saitenka (再点火) draws a Yomitan-style study overlay directly onto mpv's video: Japanese subtitles colored by what your Anki/FSRS memory already knows, a hover → multi-dictionary tooltip with pitch and frequency, and one-key sentence mining into Anki — without ever leaving the player.

It's local-first: readings and pitch always come from your dictionaries, never from a language model. It runs on Python 3.13+, including the free-threaded (no-GIL) build.

Hover tooltip over a colored subtitle

What it does¶

FSRS subtitle coloring

Every word is colored by how well you know it — driven by your real Anki/FSRS state — so you see what's new at a glance instead of re-reading what you've mastered.

Features
Hover → dictionary tooltip

Point at a word for a Yomitan-like popup: multiple dictionaries, pitch accent, frequency, and the de-inflected form — composited into mpv's OSD, no second window.

Features
One-key Anki mining

A single keypress turns the current line into a card — sentence, audio, screenshot, definition — straight into Anki over AnkiConnect.

Quickstart
Fast enough to live in your player

Warm hover in ~0.5 ms, cold first paint ~22 ms (on an M3 Pro). It stays out of the way of playback.

Benchmarks

Get started¶

Saitenka installs as a uv tool:

uv tool install "saitenka[full]"
saitenka setup   # installs mpv+ffmpeg, writes config, imports dictionaries
saitenka run <your-episode.mkv>

Then hover a colored word and press Ctrl+M to mine it.

Install guide Quickstart Why Saitenka?

Where to go next¶

Install and Quickstart — from zero to your first mined card.
Usage — features, keyboard shortcuts, configuration, and the CLI reference.
Why Saitenka — how it compares to SubMiner and Autocards, and how fast it is.
Contributing — architecture, development, and the changelog. Future direction lives in GitHub issues.