Native subtitles with instant dictionary definitions, on top of the videos you already watch. Open YouTube and it just works - pick your track and go. Watching anime? Find subs without leaving the page.
Then hover a word. No pausing, no tab switching, no copy-paste into a dictionary.
Suddenly you realize - watching foreign videos with English subtitles never worked. What you actually needed was native subtitles with dictionary definitions on demand. That's the whole idea ๐
Kitsumi started from one problem: "I want to study Japanese from anime - without spending 40 minutes gluing five plugins together before episode 7."
So it collapses the whole immersion workflow into the page you're already on: gets subtitles, puts them over the video, then you hover a word, understand it, and keep going.
What it does: deletes the setup tax. Searching subs, syncing them, finding dictionaries, cutting audio for a card - that's overhead, and overhead is worth automating.
What it does not do: learn the language for you. Hovering a word is recognition. Recognition feels like progress and mostly isn't. The part that builds memory is the part where you guess before you look, retell the scene, and produce the sentence yourself - Kitsumi hands you the tools for that, but you still have to do it. Anyone selling you passive fluency is selling you a fairy tale.
The deal is simple:
| Kitsumi handles | You handle |
|---|---|
| Subtitles, timing, dictionaries, lookups, cards, stats | Attention, guessing, recall, output |
It doesn't replace your course, Anki, textbooks or a tutor. It's the missing layer between "I'm watching something" and "wait - I actually learned something."
And yes, Japanese gets special treatment. Kanji declared war first. ๐ฏ๐ตโ๏ธ
- Install, open a video, click the ๐ฆ icon.
- Pick your language on the welcome screen - it sets up sane defaults.
- Download one dictionary from the Dict tab. One is enough to start.
-
Get subtitles - Extract them from player, Search from providers inside Search tab, or drop in your own
.srt/.vtt/.assin local tab. - Hover a word. That's the whole loop.
Everything below is optional. Keep settings default until you need something.
Extract subtitles from the player or load your own - .srt ยท .vtt ยท .ass - over Netflix, Crunchyroll, local files and most HTML5 players. On YouTube it works automatically - just pick the right track.
| ๐จ Appearance | โ๏ธ Control | ๐ Continuity |
|---|---|---|
| Font, size, colors, outline, position | Timing offsets down to ยฑ10 ms | Offsets saved per show |
| Resizable, movable overlay | Nudge strip: -1000 ms โ +1000 ms | Auto-next episode keeps your setup |
Because some subtitle files have perfect translation, perfect timing, and the visual design of a 2007 fansub made in Microsoft Paint. And some were synced using the method known as "rough and ready."
| ๐ฆ Jimaku | ๐ OpenSubtitles | ๐ฅ SubDL | ๐ Kitsunekko |
|---|---|---|---|
| Anime ยท ๐ฏ๐ต JP | Movies & anime ยท ๐ | Large database ยท ๐ | Anime ยท ๐ฏ๐ต JP |
No more Anime โ Google โ random subtitle site โ popup โ popup โ wrong episode โ wrong format โ why am I here. Side quest: skipped.
Subtitle text becomes an interactive dictionary: readings, definitions, part of speech, level and frequency wherever the data exists.
๐ฏ๐ต Japanese runs the full stack: Jitendex ยท JMdict ยท JMnedict ยท KANJIDIC ยท JLPT levels ยท example sentences.
๐ Everything else runs on Yomitan-format dictionaries - the same ecosystem Yomitan users already have. 46 languages ship with a ready-to-install catalog; the handful of long-tail languages with no Yomitan dictionary in existence yet are labeled as such in the list, so you're never left guessing which is which.
๐ Bring your own: .tsv, Yomitan .json, Yomitan .zip. Personal wordlists, anime terminology, domain glossaries - Kitsumi would like to meet your dictionary.
Study Mode turns the current line into a working vocabulary list. Mark words you already know as memories - they turn gray and stop drawing your eye, so what's left highlighted is what's actually new.
Later, add AnkiConnect - the moment you actually want to make flashcards for the words you meet:
| ๐ Word | ๐ Reading | ๐ก Definition | ๐ฌ Sentence | ๐ธ Screenshot | ๐ Audio | ๐ฌ Clip |
|---|---|---|---|---|---|---|
| โ | โ | โ | โ | โ | โ | โ |
Quick-mine hotkey, draft queue, configurable context. Cards are the residue of understanding something, not a substitute for it - mine the words you'd be annoyed to forget, not every word you saw. Your future self could spend hours reviewing every excess card you make now. Respect your time.
Kitsumi can show how much of a line is new to you and underline unknown words, so you can steer toward material you understand ~90-95% of - hard enough to learn from, easy enough to keep watching. That one setting will do more for you than any XP bar in this README.
AI is opt-in and dormant until you paste your own key. No key, no AI, no quiet background requests.
| Service | What it does | Setup |
|---|---|---|
| Groq | Cleaner definitions & example sentences | Your own API key |
| WaveSpeed | Generated scene imagery for cards | Your own API key |
๐ Articles Reader brings the same hover dictionary to ordinary webpages - news, Wikipedia, docs, blogs, or that Japanese site you should not have opened at 3 AM. ๐ผ๏ธ OCR (browser TextDetector or Tesseract.js) pulls text out of images and video frames. Both opt-in.
There are 36 ranks, from ๐ฑ Sprout to ๐ซ Soul of the Language, 200+ achievements, streaks, per-language profiles, and a stats page that answers "have I actually been doing this?"
Read this part honestly. XP and ranks measure activity, not ability. Minutes watched and words hovered are a log of what you did, not a certificate of what you know. They're here because a visible streak is a decent reason to open the app on a bad day - that's the whole job. Level estimates (JLPT-style where a framework exists) are rough progress signals, not certifications. Kitsumi is not secretly your examiner.
No quests, no lives, no virtual fox whose emotional wellbeing depends on your Tuesday. It records what happened while you were busy enjoying something. Then one day you open the profile and go "...huh. I'm actually pretty far."
| โ No accounts | โ No subscriptions | โ No telemetry | โ No analytics |
|---|
Your study data, mined words and settings live in your browser's local storage. Nothing is sent to a Kitsumi server, because there is no Kitsumi server.
Network requests happen only when you ask for them: searching subtitles, downloading a dictionary, or an AI feature you switched on with your own key. "No collection" is not the same as "no network", and you should be suspicious of any extension that blurs the two.
About permissions: the extension asks for broad site access because subtitles have to be drawn over players it can't predict, including ones inside iframes. If that's more than you want to grant, restrict it - right-click the icon โ This can read and change site data โ On specific sites, and allow only the streaming sites you use. Everything above still works. That's a supported way to run it, not a workaround.
Free, no subscription, no paywalled features. If Kitsumi saved you from subtitle suffering, or from handcrafting Anki cards like it's 2014 - share some love:
Bugs and feature requests go in Issues. Questions, community talks, recommendations and general fox business go to Discord.
Kitsumi - status
[ OK ] subtitle overlay ยท search ยท precision sync ยท auto-next
[ OK ] interactive dictionary ยท 46 languages ยท custom dictionaries
[ OK ] study mode ยท anki mining ยท articles reader ยท OCR
[ OK ] progression ยท statistics ยท privacy
[WARN] "just one more episode"
[WARN] user has been watching for 6 hours
[WARN] user somehow knows 1,200 new words
[WARN] kanji remain undefeated
[STATUS] ๐ฆ
"I-it's not like I built you an entire language-learning ecosystem disguised as a subtitle extension or anything..." ...baka. ๐ฆโจ