Skip to content

How it compares

Saitenka, SubMiner, Autocards, and Anki Miner all put Japanese vocabulary into Anki, but from four different angles.

  • Saitenka is a grounded, FSRS-driven engine composited inside mpv — colored subtitles, a hover dictionary tooltip, and one-key mining, all drawn into mpv's own video surface.
  • SubMiner is a feature-broad Electron app with a large integration surface and turn-key desktop installers (AniList scrobbling, Jellyfin integration, a media launcher, cross-machine sync).
  • Autocards is a retroactive media back-filler: it doesn't create cards or show dictionaries — it batch-attaches a screenshot and sentence audio to text-only cards you made elsewhere, by matching each card's Sentence field against the subtitle's timing.
  • Anki Miner is a batch-mining desktop GUI: point it at a folder, whole series, YouTube playlist, audiobook, manga, or ebook, review the candidate words in a curator, and it builds a frequency-ranked deck in one pass — the opposite pole from Saitenka's live, one-word-at-a-time overlay.

So this is a map of trade-offs across different jobs, not a scoreboard. For raw rendering and lookup latency, see Speed; to try it, see Get started.

Capability matrix

Capability 再点火 Saitenka SubMiner Autocards Anki Miner
Category live in-mpv engine Electron mining app retroactive enricher batch mining GUI
Runs inside mpv's own surface — no second window, fullscreen/airspace-safe ✅ ❌ ❌ ❌
Multi-dictionary Yomitan tooltip · pitch accent · frequency ✅ ✅ ❌ ✅
One-key + bulk Anki mining (audio, screenshot, reading, freq) ✅ ✅ ❌ ✅
Retroactive media back-fill onto existing text-only cards (sentence↔sub-timing match) ❌* ❌ ✅ ❌
Reading-aware known-word matching (homograph-safe) ✅ ✅ ❌ ❌
N+1 sentence targeting ✅ ✅ ❌ ✅
Live FSRS review-state coloring — forgotten words resurface ✅ ❌ ❌ ❌
Grounded / local-first (readings never from an LLM) ✅ ✅ ✅ ✅
Batch folder / whole-series frequency-ranked deck building ❌ ❌ ❌ ✅
Reading mining — manga (mokuro) · novels (epub/txt) · pasted text ❌ ❌ ❌ ✅
Audiobook / YouTube URL + playlist mining (yt-dlp) ❌ ❌ ❌ ✅
Pre-mine word curation — review candidates before cards ❌ ❌ ❌ ✅
jimaku.cc · TsukiHime subtitle fetch ✅ ✅ ❌ ❌
Built-in subtitle retiming/resync (alass) ✅ automatic ⚠️ ❌ external tool ✅ utility
Local subtitle generation (Whisper ASR, when no subs exist) ❌ ❌ ❌ ✅
Extra subtitle sources (Animetosho) · YouTube subs ❌* ✅ ❌ ⚠️ YouTube only
AniList progress scrobbling ❌ ✅ ❌ ❌
Jellyfin integration · media launcher (fzf/rofi) ❌ ✅ ❌ ❌
Episode analysis · local session history ✅ ✅ ❌ ✅
Cross-machine stats/history sync ❌ ✅ ❌ ❌
Mined-audio loudness normalization ❌* ✅ ❌ ❌
Built-in latency profiling / OpenTelemetry traces ✅ ❌ ❌ ❌
Free-threaded parallel rendering (Python 3.14t) ✅ ❌ ❌ ❌
One-command installer / portable bundle ✅ ✅ ✅ ✅
Native desktop packages (AppImage · DMG · AUR · winget) ❌ ✅ ❌ ⚠️ AppImage · deb · exe
Core license Apache-2.0 GPL-3.0 GPL-3.0 GPL-3.0
Linux · macOS · Windows ✅ ✅ ❌ Windows ✅

✅ yes · ❌ no / out of scope · ⚠️ partial · * on the roadmap

Where it fits in your setup

Saitenka is the study layer at the mpv playback point. It doesn't acquire, organize, serve, or track your library — it composes with the tools that do, rather than replacing them. If you already run a media-server rig, this is the division of labor:

Pipeline stage Common tools Saitenka
Acquire episodes Sonarr (PVR — monitors RSS, grabs + renames) · Taiga (RSS/torrent feeds, Windows) · Usenet/torrent clients —
Identify & organize the files Shoko (AniDB hashing) + Shokofin —
Serve & stream the library Jellyfin · Plex · Emby —
Synchronize a watch party Syncplay (coordinates each friend's local player) ✅ runs alongside
Play the file mpv ✅ attaches here
Color words · dictionary · mine Saitenka ✅ its whole job
Spaced repetition Anki + FSRS ✅ mines in via AnkiConnect
Track list · discover · scrobble Taiga (Windows) · Trackma (cross-platform) runs alongside

The one hard rule: Saitenka draws into a real, local mpv process — over its IPC socket, with the auto-start plugin. It rides the player, not the server, so it works with a Jellyfin/Shoko library only when you play the file in mpv. Watching inside a Jellyfin/Plex client — web, TV, or phone — is out of reach, because those aren't mpv.

The short version

  • Reach for SubMiner if you want the widest set of turn-key integrations and a packaged desktop app (AniList scrobbling, Jellyfin, a media launcher, cross-machine sync, native installers).
  • Reach for Autocards if you make cards fast in a browser texthooker and want to attach the media in one pass afterward (Windows-only).
  • Reach for Anki Miner if you want to batch-mine whole series, manga, ebooks, audiobooks, or YouTube into a frequency-ranked deck, curating the candidate words before cards are made — a tabbed desktop app, not a live overlay.
  • Reach for Saitenka if you want a fast, single-surface, FSRS-grounded engine that draws straight into mpv — no second window, forgotten words resurfacing from your review state.

The workflows compose — Autocards' back-fill is on Saitenka's roadmap as an in-engine step.

Visual parity harness

Comparison isn't only about features. Saitenka ships a visual parity harness in overlay/compare/ that renders its in-mpv tooltip side-by-side with real Yomitan and SubMiner popups for the same words, so its dictionary UI can be checked against the tools it reproduces. (No screenshots here yet — the harness generates them locally.)