A video player for learning Japanese from the anime and series you watch.
Click a word in the subtitles, read it in your Yomitan dictionaries with pitch accent and frequencies, and make an Anki card with the sentence, the frame and the audio of the line in one click. Written in Rust, with mpv for playback.
-1e2a44?style=flat-square&logo=linux&logoColor=white){kind=icon}
A click on 食べ in 食べました: the popup finds 食べる through the conjugation, with its frequency, pitch accent and meaning. Project-made video and test dictionaries.
Screenshots · Why Nikomiru · What works today · Quick start · Architecture · Roadmap · Downloads · License and credits
 The card is in Anki: overwrite it, open it in Anki or add another. |
 The subtitle list: search, jump to a line, click its words too. |
 Subtitles found online, the Japanese ones of the episode first. |
 Settings: the subtitle services to use, in your order. |
 Settings: the mpv options as mpv.conf text and as controls. |
 Settings: the Anki deck, note type and fields (Lapis ready). |
- Every word is clickable. Nikomiru draws the subtitles itself (SRT, ASS, WebVTT, embedded tracks), so any character can be looked up; mpv draws only the video.
- Your Yomitan dictionaries. Import the same ZIP files as Yomitan: terms, kanji, frequencies and pitch accent, with conjugations undone as Yomitan does.
- One-click Anki cards. A Lapis card with the word in the sentence, the frame, the audio of the line and a link that reopens the scene; Ctrl+Enter from the keyboard. No Anki? The cards can stay in Nikomiru and come back on its own review.
- Subtitles that fit. Files next to the video are found also by episode; the missing ones come from Jimaku or OpenSubtitles, by hand or by themselves; an offset is found from the audio; a second subtitle (a translation) can sit above the first.
- mpv as you like it. Your
mpv.confloads with a preview of every line, presets switch in one click, and the options apply without reloading the file. - Private by default. Nothing goes online unless you ask; keys stay in a file only you can read, never in the logs.
Version 2.10.1 is for Linux. The status of each area follows the ROADMAP.
| Area | Status |
|---|---|
| Player: open, play, seek, speed, tracks, chapters, fullscreen, resume, screenshots | Works; the manual checklist on Wayland and X11 is still open (M1) |
| Subtitles: SRT, ASS, WebVTT, embedded tracks, encodings, list, blur, offset | Works (M2) |
| Dictionaries: Yomitan import, lookup with deinflection, popup with pitch accent and frequencies | Works with the project's test dictionaries; the comparison with Fushi on real dictionaries is still open (M3) |
| Mining to Anki: Lapis cards with sentence audio, frame and a link back to the scene | Works against a simulated AnkiConnect; not yet checked with a real Anki and AnkiDroid (M4) |
| Online subtitles: Jimaku and OpenSubtitles, by hand or automatic; alignment with the audio; secondary subtitle | Works against simulated services; not yet checked with real keys (M4b) |
mpv options: controls, mpv.conf presets, import and export |
Works (M4b) |
Library, word audio (also from a local android.db) |
Works with project-made files; a real library of hundreds of episodes is still to be checked by hand (M5) |
| Statistics, the full settings (themes, languages, shortcuts), cards made by hand, the keyboard alone | Works (M6) |
| Series identified online (AniList, Jikan); Jimaku by AniList id | Works against simulated services and the real AniList (M7) |
| Media servers: Jellyfin and Plex, in a tab of their own, played with subtitles, lookups and cards | Works with a real Jellyfin; Plex only against a simulated server (M11) |
| Backup to a file, sync of progress, statistics and cards through WebDAV | Works with a real Nextcloud (M12) |
| Advanced ASS, screenshots with subtitles | Works; real typeset fansubs still to compare with libass (M13) |
| Texthooker for visual novels, with lookups and cards of text only | Works with a simulated hooker (M14) |
| StarDict, DSL and MDict dictionaries, their definitions as plain text | StarDict works with a real dictionary; DSL and MDict only on files made by the tests (M15) |
| Books: EPUB with furigana and pictures, vertical or horizontal, lookups and cards of text only | Works with a real EPUB; a book with furigana still to check (M16) |
| Audiobooks: sound with timed text (LRC, SRT, VTT, ASS), read along in the list or in their EPUB, cards with the sound of the line | Works with a file made by the tests; a real audiobook still to check (M16) |
| Manga: CBZ, a balloon read on this computer by a model fetched on request, lookups, cards with the part of the page | Works; the real model was tried on computer-set text, real pages still to check (M16) |
| Words in the subtitles: the words already in Anki underlined, readings over the kanji, ← → word by word | Works with the project's test dictionaries; to check with JMdict (M9) |
| Moving pictures: a card's picture as the animated sentence (WebP, AVIF), thumbnails over the progress bar | Works against a simulated AnkiConnect; to check in Anki and AnkiDroid (M10) |
| Web pages read as books: an address typed on the Books page, or an HTML file, with lookups and cards | Works with pages made by the tests; real sites still to check (M18) |
| Review in Nikomiru: cards kept without Anki, brought back with FSRS on the Library page, synced with their media, with statistics | Works; the sync of the review against the simulated WebDAV only (M17) |
| Packages (AUR, Flatpak) | Built and checked; publishing on the AUR and Flathub follows (M8) |
The user guide (in Italian) covers everything below in more detail. On Arch Linux, packaging/aur/PKGBUILD builds a package (makepkg -si).
You need Rust through rustup (the version is pinned in rust-toolchain.toml) and these system libraries: libmpv ≥ 0.37, FFmpeg ≥ 6, OpenGL with EGL (Mesa or NVIDIA's driver), the Wayland or X11 client libraries and libxkbcommon, fontconfig, OpenSSL 3 and a font with Japanese glyphs (for example Noto Sans CJK). Building also needs the development files of libmpv and OpenSSL, and pkg-config.
git clone https://github.com/KvasFan/nikomiru.git
cd nikomiru
cargo xtask install # release build into ~/.local/bin, with the desktop entry
nikomiru path/to/episode.mkvcargo xtask install also registers Nikomiru for nikomiru:// links, so the link in a card opens its scene. To try it without installing: cargo run --release -p nikomiru-app -- path/to/episode.mkv.
Then:
- Dictionaries: 📚 in the bar → import your Yomitan dictionaries (JMdict, a frequency and a pitch accent dictionary). Nikomiru ships none.
- Anki: install the AnkiConnect add-on in Anki and keep Anki open. In Nikomiru, ⚙ → Anki → "Create Lapis deck", or choose your own deck and note type.
- Online subtitles (optional): ⚙ → Subtitle services → tick the services, paste a free Jimaku key and/or an OpenSubtitles consumer key.
Logs are in ~/.local/state/nikomiru/logs/.
A Cargo workspace of 11 crates: pure logic (nikomiru-core, -text, -dict, -subs, -anki, -library, -online), the programs it drives (nikomiru-media for ffmpeg, nikomiru-player for libmpv), the egui interface (nikomiru-ui) and the binary (nikomiru-app). mpv draws the video with its OpenGL render API into the window under the interface; workers on their own threads talk to Anki, the online services and ffmpeg, so the interface never waits. See SPECIFICATION §5 and the architecture decisions.
| Document | Contents |
|---|---|
| docs/SPECIFICATION.md | Purpose, licenses, architecture, requirements, decisions and risks |
| docs/ROADMAP.md | Milestones M0–M8 up to 1.0, exit criteria, post-1.0 backlog |
| docs/TEST.md | Test strategy, fixtures, conformance with Fushi, case catalog |
| docs/adr/ | Architecture decisions with the spike measurements |
cargo install --locked cargo-deny
cargo install --locked cargo-about --features cli| Command | What it does |
|---|---|
cargo xtask ci |
fmt, clippy -D warnings, architecture checks, interface strings, licenses (cargo deny), traceability, tests |
cargo xtask ci --full |
as above, plus fixtures, L4/L5 tests, the offline build, the release build, the GUI end-to-end tests (L6, in an off-screen kwin_wayland on a private dbus-daemon; the accessibility test also needs at-spi2-core) and THIRD_PARTY_LICENSES.html |
cargo xtask fixtures [--large] [--perf] |
generates test media and dictionaries in target/fixtures/ (requires ffmpeg) |
cargo xtask trace |
requirements ↔ tests ↔ milestones matrix in target/trace.md |
cargo xtask about |
generates target/THIRD_PARTY_LICENSES.html |
cargo xtask coverage |
line coverage of the logic crates with cargo-llvm-cov, at least 80% each |
cargo xtask flatpak |
the Flatpak of the last commit (packaging/flatpak), into target/nikomiru.flatpak |
cargo xtask check-i18n |
every interface string used and translated, no untranslated text in the views |
cargo xtask install [--prefix DIR] |
builds the release binary into DIR/bin (default ~/.local) and installs the desktop entry |
The screenshots on this page come from the GUI harness: NIKOMIRU_SCREENSHOTS=<folder> cargo test -p nikomiru-app --features test-harness --test e2e -- --include-ignored screenshots_of_the_interface.
0.1.0 brought mining to Anki (M4) and online subtitles with the mpv presets (M4b), 0.2.0 the library and word audio (M5), 0.3.0 statistics, the full settings, cards made by hand and the keyboard alone (M6), 0.4.0 series identified online (M7), 1.0.0 stabilization and packages (M8), with the words in the subtitles (M9) and the moving pictures (M10), 1.5.0 media servers, backup and sync, advanced ASS, a texthooker, more dictionary formats and books (M11–M15, the books of M16); what is left is the checks by hand of each milestone. 1.6.0 brought vertical writing and pictures to books, 1.7.0 audiobooks, 2.0.0 manga (D-12, D-28), 2.1.0 video cards, 2.2.0 the interface in Japanese, 2.3.0 keys in the system keyring, 2.4.0 a review of its own (M17), 2.5.0 the review in the sync, 2.6.0 web pages read as books (M18), 2.7.0 the review's statistics, 2.8.0 its cards listed, 2.9.0 audiobooks read along in their book, 2.10.0 a look after GNOME's Adwaita. Builds for Windows, macOS and Android are planned (D-16), after a study of what each takes. The details are in the ROADMAP.
| Format | Status |
|---|---|
Source (cargo xtask install) |
Available from 0.1.0 |
| PKGBUILD (AUR) | In packaging/aur from 1.0.0; on the AUR soon (D-5) |
| Flatpak (freedesktop runtime) | cargo xtask flatpak from 1.0.0; on Flathub soon (D-5) |
GPL-3.0-only. Nikomiru is a Rust rewrite of Fushi and derives from GPL-3.0 code: Fushi, Yomitan (deinflection tables), hoshidicts and the Lapis note type, whose templates ship unchanged in assets/anki/lapis/. The logo and icons are original works of the project, under the same license.
Reading manga uses manga-ocr by kha-white (Apache-2.0), a model trained on the Manga109-s dataset, as an ONNX export. It is not part of Nikomiru: it is fetched only if you ask for it.
The allowed dependencies are MIT, Apache-2.0, GPL and LGPL, with a few permissive exceptions documented in SPECIFICATION §3 and checked by cargo deny; cargo xtask about lists them all in THIRD_PARTY_LICENSES.html.
Nikomiru was called Anikoto until 2026-10-19 (decision D-11).