The FlixFluent vocabulary trainer

At a glance

• •Included in the standard $17 / month FlixFluent subscription (web app)
• •FSRS scheduling via the open-source rs-fsrs implementation, 90% retention target
• •Cards auto-created from words you look up while watching, with source sentence and episode reference
• •Anki .apkg and CSV import; CSV export
• •Adaptive pacing: 5–20 new cards/day based on your trailing 14-day retention
• •Receptive → productive graduation; optional typed recall (off by default) with near-miss tolerance
• •No self-run peer-reviewed studies; design follows published spacing / retrieval-practice research

What is the vocabulary trainer?

A flashcard review system that converts your real watching activity into a daily review queue, so the words you actually met in shows are the words you retain.

It lives in the FlixFluent web app at app.flixfluent.com and is included in the normal $17 / month subscription — there is no separate purchase. The core loop: watch with the extension, click words you don't know, and the trainer collects those words into a review pool. Each day it deals you a queue of due reviews plus a measured number of new cards, scheduled by FSRS.

The trainer is deliberately biased toward being sustainable. Daily new-card volume adapts to your actual measured retention, cards from the same sentence never appear on the same day, and a mistyped answer that is one letter off is treated as a near miss rather than a failure. The goal is real retention without the review-debt spiral that makes people abandon flashcard systems.

Where do cards come from?

Two sources: automatically from the words you look up while watching, and manually via Anki or CSV import.

Organic capture: when you click a word on Netflix or YouTube, the trainer records the dictionary form, the exact inflected form you saw, the full subtitle sentence it appeared in, and which episode it came from. The stored sentence becomes the card's context — productive cards test you with a cloze (fill-in-the-blank) built from the very sentence you originally watched, and the episode reference means a card can tell you where you met the word.

Import: the trainer accepts Anki deck files (.apkg) and CSV uploads, so an existing collection can move in without retyping. Your collection can be exported back out as CSV at any time (there is no .apkg export). Imported cards carry their own gloss and are flagged as imported so they can be managed separately.

How is scheduling done? (FSRS)

Scheduling uses FSRS — Free Spaced Repetition Scheduler — via the open-source rs-fsrs Rust implementation, with the request-retention target fixed at 0.90.

FSRS models each card with three quantities: difficulty, stability (how long the memory lasts), and retrievability (the probability you can recall it right now). After every review it updates the card's stability and difficulty and schedules the next review for the moment predicted retrievability falls to the retention target — for FlixFluent, 90%. This is the same algorithm family that modern Anki ships as its recommended scheduler.

In the open srs-benchmark maintained by the FSRS authors — 9,999 anonymised Anki collections, roughly 350 million real reviews — FSRS variants predict recall substantially more accurately than the classic SM-2 heuristic that older flashcard systems use; the benchmark's summary page reports the current FSRS release beating SM-2 for over 99% of users on log loss (with the stated caveat that SM-2 was never designed to output probabilities). FlixFluent uses the stock FSRS parameters; we do not currently train per-user weights.

The 90% target is a deliberate trade-off: pushing target retention higher makes reviews balloon for small retention gains, and lower targets forget too much. It is fixed rather than user-configurable in the current version.

What does the trainer add on top of FSRS?

FSRS decides when a card comes back. Everything around that — how many new cards you get, what order, what counts as an answer — is FlixFluent's own layer, tuned to keep daily load humane.

Two-button grading with automatic Easy: reviews are graded with just Again and Good (swipe or click). If you answer Good within about 3 seconds, the trainer upgrades the grade to Easy automatically — near-instant recognition is treated as fluent knowledge, a heuristic drawn from Anki community practice. This keeps grading friction near zero without collapsing FSRS's rating scale to two levels.

Retention-adaptive pacing: the daily new-card budget starts at 10 and moves between 5 and 20 based on your trailing 14-day retention — below 85% retention it drops to 5, at 92%+ with a light due load it rises to 15, at 95%+ with a very light load it rises to 20, and it never exceeds 20. The budget is further trimmed or expanded (×0.7–1.1) by how hard the upcoming words look: their corpus frequency rank and the trailing average FSRS difficulty of your due cards. If you want more, a "show me 5 more" button adds new cards in explicit increments of five.

Same-sentence burying: grading a card buries every other card that shares its source sentence until the next day, so one sentence never yields three near-duplicate reviews in a session. New cards are interleaved into reviews at roughly 1 new per 3 due; when a backlog has built up, due reviews are front-loaded first. Days roll over at your local midnight, not UTC.

Small quality-of-life mechanics: the last review can be undone, any card can be ignored permanently (the replacement for classic leech auto-suspension), and per-card statistics feed a visible "mastered" count (cards whose memory stability exceeds three weeks) plus a 30-day retention figure.

Receptive and productive recall

Cards start receptive (see the word, recall the meaning) and graduate to productive (see the meaning, produce the word) once the receptive direction is established.

Recognising a word and being able to produce it are different skills, and productive recall is the harder, higher-value one. In the default "auto" mode, a card is promoted to productive practice once it is genuinely established — at minimum three successful reviews in review state, plus stability thresholds (roughly one to two weeks of demonstrated memory, sooner for cards you find easy and have never lapsed on). Productive cards present the card's original sentence with the word blanked out.

The direction is user-configurable: you can pin the trainer to receptive-only, productive-only, or leave it on auto-graduation.

Typed recall, near misses, and multiple choice

Productive cards can require actually typing the answer — with an on-screen keyboard for non-Latin scripts and a forgiving grader that distinguishes typos from wrong answers.

Typing is the strictest form of retrieval practice, and the trainer supports it natively: an optional "force me to type" mode (off by default — you can enable it in settings) requires a typed answer on productive cards instead of a reveal-and-swipe. For scripts your physical keyboard may not cover, the trainer ships on-screen keyboards, including Korean (with jamo composition), Arabic, Hebrew, and Hindi.

The grader normalises the answer (case, accents, and for Korean the dictionary-form ending) and then allows a small edit distance — one character for Korean / Chinese / Japanese scripts, two for others. An answer within that distance is classified as a near miss: you see the correction and continue, rather than having the card fail outright. Genuine failures still grade Again.

There is also a multiple-choice mode: the correct answer plus three distractors drawn from your own card pool and length-matched so the right answer is not guessable by shape. Audio is covered too — cards can speak the word via text-to-speech, with an auto-play option.

All of this is user-configurable in the trainer settings: typing on or off, recall direction, sound auto-play.

What research is the design based on?

The trainer's mechanics map onto three well-replicated findings in the memory literature: the spacing effect, the testing effect, and incidental vocabulary learning from video.

Spacing: distributing reviews over increasing intervals produces more durable memory than massed repetition — the meta-analytic evidence covers hundreds of experiments (Cepeda, Pashler, Vul, Wixted & Rohrer, 2006, Psychological Bulletin). Spaced repetition scheduling is the entire point of FSRS, and its accuracy at predicting recall is measurable on public data (see the srs-benchmark below).

Retrieval practice: actively recalling an item strengthens memory more than re-reading it, especially at a delay (Roediger & Karpicke, 2006, Psychological Science). This is why the trainer is recall-first — reveal-then-grade at minimum, typed production at the strictest setting — rather than a re-exposure tool.

Learning vocabulary from television: incidental vocabulary acquisition through watching L2 television is real and measurable, and is strongly affected by frequency of encounter and prior knowledge (Peters & Webb, 2018, Studies in Second Language Acquisition). The trainer closes the loop the research points at: the words you met on screen get the repeated, spaced encounters that watching alone cannot guarantee.

Honest limits: FlixFluent has not run its own peer-reviewed efficacy studies, and we make no numeric "learn N× faster" claims. What we can state factually is what the system does: FSRS scheduling at a 90% retention target, recall-based review, and pacing that adapts to your measured retention — with the supporting literature linked below.

Frequently asked questions

Is the vocabulary trainer included in the FlixFluent subscription?

Yes. It is part of the standard $17 / month subscription, in the web app at app.flixfluent.com. There is no separate purchase.

Which algorithm does the trainer use?

FSRS, via the open-source rs-fsrs Rust implementation, with stock parameters and a fixed 90% retention target.

Do I have to type answers?

No. Typed recall ("force me to type") is optional and off by default. When enabled, near misses within one or two characters are corrected rather than failed.

Can I import my Anki deck?

Yes — .apkg files import directly, as do CSV files. Export is CSV only; there is no .apkg export.

How many new cards do I get per day?

Base 10, adapting between 5 and 20 from your trailing 14-day retention and how difficult the upcoming words look. A "show me 5 more" button adds more on demand.

Why only Again and Good buttons?

Two buttons keep grading fast and consistent. A Good answered within about 3 seconds is automatically upgraded to Easy, so FSRS still receives more than two rating levels.

Does the trainer work if I don't watch Netflix or YouTube?

Yes, with imported cards (Anki / CSV). But its distinguishing value is turning your own watch-time lookups into cards with their original sentence and episode context.

Can I review on my phone?

The trainer is a web app and opens in mobile browsers; there is no native mobile app.

What happens to cards I keep failing?

Nothing is silently suspended. You can ignore any card permanently, undo your last review, and same-sentence siblings are buried until the next day to avoid pile-ups.

Is there proof this makes you learn faster?

We make no speed claims. Spacing and retrieval practice are among the best-replicated effects in memory research (sources below), FSRS's scheduling accuracy is measured on ~350 million public reviews, and the trainer implements those mechanisms — but FlixFluent has not run peer-reviewed studies on its own product.

Related pages

• About FlixFluent
• FAQ
• Glossary — SRS, FSRS, retrieval practice
• Learn Korean with Netflix

Sources & further reading

• FlixFluent homepage
• FlixFluent on the Chrome Web Store
• FlixFluent pricing
• Cepeda, Pashler, Vul, Wixted & Rohrer (2006) — Distributed practice in verbal recall tasks, Psychological Bulletin 132(3)
• Roediger & Karpicke (2006) — Test-enhanced learning: taking memory tests improves long-term retention, Psychological Science 17(3)
• Peters & Webb (2018) — Incidental vocabulary acquisition through viewing L2 television, Studies in Second Language Acquisition 40(3)
• rs-fsrs — open-source Rust implementation of the FSRS scheduler
• srs-benchmark — open benchmark of spaced-repetition algorithms (9,999 collections, ~350M reviews)
• Benchmark of spaced repetition algorithms — summary page (FSRS vs SM-2)