# 「Looks like she's my woman now」: a Korean indie developer turned an anime waifu into a Japanese tutor · Mobile app

> Source: https://stringers.io/en/apps/ios_6828335335.html (Stringers / 什么值得做, generated 2026-11-20)
> Stringers tracks which small products suddenly took off each day, investigates why one by one, and distils playbooks from them. Numbers come from 15 databases we collect daily, each with source and date.
> For agents: these investigations and the underlying databases (verified revenue, App Store / Google Play charts, ad libraries, GitHub, Product Hunt…) are queryable live over MCP — tools feed (daily breakouts) / plays (playbooks) / hot / products / data. Setup: https://github.com/komako-workshop/stringers (the user signs in with Google at stringers.io/en/start.html to get a personal URL). Site index: https://stringers.io/llms.txt

> It hasn't really taken off yet. It rode Korea's low-bar education chart and comments on Instagram that got seen: a wave of anime fans flooded in over a weekend writing "this is my wife," pushing it to #51 in the Education Free chart.

[Mobile App](https://stringers.io/en/apps/index.md) · iOS · written 11-17

하나비 - 유이와 일본어 회화(Hanabi) · YongJin Kwag

# 「Looks like she's my woman now」: a Korean indie developer turned an anime waifu into a Japanese tutor

**What it is** An AI Japanese conversation app for Korean anime fans: you speak Japanese by voice with Yui, the only Live2D fox-spirit girl, with furigana, translations, corrections, and lessons running from the kana charts to travel conversation. The higher the affection, the more lines unlock. Free for 5 minutes a day, with a non-renewing 30-day pass and gems for purchase.

**Why it took off** It hasn't really taken off yet. It rode a low bar on the Korean education charts and comments on Instagram that got seen. A wave of anime users rushed in over one weekend writing 「this is my wife」, which pushed it to #51 on the education free chart.

**B** **How you could follow it** Take the formula of one fixed character + lessons + mobile-game-style progression and per-minute pricing, and apply it to other subculture × language pairs, like anime fans learning Japanese in the English-speaking world or K-pop fans learning Korean. Spend real money on the character's art and voice, and go after the comments under viral short videos in that niche, letting the character ride current events.

Grade B: the formula is clear and almost nobody knows about it, and the demand has been validated in Korea separately by anime Japanese web courses and AI character chat. But 하나비 itself has only a dozen-odd ratings, so it hasn't proven the combination can scale, and character art and voice are a real barrier. Few people know about it yet

Est. monthly revenue (App Store) **under $1K** We estimated this from its chart ranks and rating growth across storefronts, calibrated against verified revenue; typical error is 2–3x. Read it as a tier, not an exact number.

Data (updated daily) · First launched 2026-11-06 · Korean rating 18 · 11-20 charted in 1 stores

iOS · best category chart rank · by store

Data 2026-11-16 → 2026-11-20 (5 days, 5 observations). Lower rank is higher; gaps mean it wasn't charted that day

iOS · rating count · Korea

Data 2026-11-14 → 2026-11-20 (7 days, 6 observations). Apple doesn't publish download counts; rating growth is a proxy for downloads.

하나비 (Hanabi) launched on the Korean App Store in Nov 6. 12 days later it had only 16 ratings and 31 Android installs, yet it had already squeezed into the top 60 of the Korean education free chart. The numbers are too small to call it a hit. What's worth watching is the combination: a fixed anime heroine, a course running from the kana to travel conversation, and the affection and gem system from mobile games. Almost nobody in the comments praises the lessons; they praise the character.

### A weekend of comments

On November 14 and 15, a wave of five-star reviews landed on 하나비's App Store page. The nicknames were blunt: 「하나비남친」 (Hanabi boyfriend), 「미소녀다이스키」 (I love beautiful girls). Titles read 「아무래도 내여자인것같다」 (looks like she's my woman now) and 「내 아내임」 (this is my wife).

The reviewer who wrote 「this is my wife」 also listed a set of requests:

> I'd like it to work like Grok companion, where new outfits unlock at certain affection levels. When I renew my subscription, I'd also like to hear a thank-you from her, like a VTuber, or get some affection rewards. And add a dating scenario to the practice conversations ^q^

Another user, calling themselves 「씹덕」 (a hardcore otaku), wrote: 「It really feels like I'm having a conversation, that's great. Please make an English version too.」

By 11-17, iOS had gained 14 ratings in a day, the total going from 2 to 16, with an average of 4.9. It ranked #51 on the Korean education free chart and #159 on the education top grossing chart (App Store Korea, 11-17). It has charted only in Korea.

### What it is

The app has just one character: 유이 (Yui), the fox-spirit girl from 「하나비 神社」, with Live2D character art that changes expression when tapped. The official site makes it clear that you can't choose a character.

You use it by talking to Yui in Japanese. You can type or speak, and she answers by voice. Every line she says comes with kana readings and a Korean translation, and if you get stuck you can tap a suggested reply. Your sentences are rewritten into natural Japanese, and you get a score when you finish. Alongside is a lesson track: kana traced stroke by stroke, graded vocabulary, then scenario levels like the convenience store, café, and asking for directions. Expressions you learn today get used by Yui in conversation that same day.

The progression system is all there: affection level-ups unlock new lines and voice clips, plus Yui side stories, daily missions, and learning streaks. If you buy the top two pass tiers, Yui remembers what you've talked about.

The opening of the store description is in the words the target users would use themselves: 「I want to watch anime without subtitles and understand the lyrics of songs I love, so I started learning Japanese, and got stuck at the vocabulary list」. The 「good for these people」 section lists three types: people who've deleted several vocab apps, people who feel pressure practicing with real humans, and people who love anime, manga, and Japanese music.

The developer is a sole proprietor, 곽용진 (YongJin Kwag), registered in Icheon, Gyeonggi-do; on Google Play he uses the name Idol Attention. His GitHub organization was created in June 2026, which suggests about 5 months of work (inferred). In 10 days after launch he shipped 4 versions, fixing reconnection, keyboard, and Google login, and added a home-screen prompt right after a level-up.

### How it makes money: like a mobile game, not a study app

Free practice conversation with Yui for 5 minutes a day, after which you pay. It doesn't sell a weekly or yearly subscription; it sells two things Korean mobile-game players know well (the Korean App Store in-app purchase list):

| What's sold | Pricing | What you get |
|---|---|---|
| 30-day pass, no auto-renewal | ₩4,900 / 9,900 / 19,900 / 39,900 | 10 / 20 / 40 minutes a day / unlimited; Yui remembers you on the top two tiers |
| Gems (add conversation time) | ₩1,200 / 6,000 / 12,000 / 25,000 / 59,000 / 119,000 | 150 to 23,800 pieces |

These six gem tiers match the top-up tiers of Korean gacha games exactly. The 30-day non-renewing pass is the mobile game's 「monthly card」, so users don't have to worry about being auto-charged. Selling by the minute also makes sense: real-time voice plus a large model means costs scale with call time, so pricing by time locks in margin; the ₩119,000 tier is reserved for heavy users.

On day 10 after launch it entered the Korean education top grossing chart (#126), which shows people really are paying. But the bar to get onto the Korean education top grossing chart beyond the top hundred is low, a few purchases are enough, so revenue is probably in the range of tens of thousands of won a day (estimate, unverified).

### Where users come from

It spends almost nothing, relying on Instagram.

**Commenting under other people's viral posts.** The same line 「일본어 회화 앱 하나비 유이와 실시간 대화로 연습합니다 링크는 프로필에 #서브컬처 #하나비」 (Japanese conversation app 하나비: practice with Yui in real-time conversation, link in profile #subculture #하나비) appeared under older viral Reels from several accounts: one was the Japanese creator 도도토's 「the best AI for Japanese」, one was SOOP streamer 쿠빈's 「after learning Japanese from anime」, and a few were general chat accounts unrelated to Japanese.

**Letting the character ride current events.** Before the LCK team's World Championship final, @hanabiappkr posted a Reel: fox-spirit Yui prays for the LCK to win, and along the way teaches 「神頼み」 「必勝祈願」 「がんばれ」, with 79 likes. Another VTuber-style demo, with facial capture from Live2D, has 43 likes, and the pinned comment reads 「구라 안까고 이번 기말 1등급 맞아옴」 (no lie, I got a 1 grade on this finals).

**Everything else is small.** Three YouTube Shorts, the top with 1,400 views and 23 subscribers; the X account has 62 followers. It made a separate short link for each of App Store, Google Play, and X, so how many people each channel brought is tracked separately. No ads found (the Meta ad library wasn't fully checked, and nothing was found on TikTok, so it can't be ruled out).

How the 11-17 14 ratings came about is unclear. The reviews are concentrated on 11-14 and 15, and the nickname style is very uniform, as if people from one community showed up together. But there's also very specific usage feedback: 「Yui talks too fast on hard mode」 「please support iPad」 「the screen overlaps when the foldable is closed」, which doesn't read like filler.

### Why the combination makes sense

Broken down, each piece of demand has been validated in Korea by someone:

- **Anime audiences will pay to learn Japanese.** 도도토's course 「오레노 니홍고」 (My Japanese) is pitched as 「Japanese lessons only for otaku」, and has publicly said revenue passed ₩50M, with its website claiming ₩100M. Its audience is exactly the same as 하나비's, and 하나비 has left comments under its Reels.

- **Korean young people love chatting with AI characters.** 제타 (zeta) has over 70k ratings on the Korean App Store.

- **Koreans will pay for AI speaking practice.** Speak ranked #2 on the Korean education top grossing chart, and 헤이링 ranked #6 (App Store Korea, 11-18).

하나비 sits at the intersection of these three. Compared with AI Japanese conversation tools, it sells a relationship with a character rather than practice sessions; compared with pure AI character chat, it adds lessons and corrections, plus a respectable label: I'm studying Japanese. That label lets users convince themselves and feel comfortable showing it off publicly, like the one with the 1 grade on finals.

### Who it's competing with

| Product | Status | Compared with 하나비 |
|---|---|---|
| Amiko 아미코 | 2025-10 launched, 64 ratings on the iOS Korea store, 351 on the Japan store | The closest one, also practicing Japanese with an AI character; but it has multiple characters and 15 languages, and works through scenario role-play |
| 일본어 회화 도우미 아오이 | About 2022 years old, 44 ratings on the Korean store | An AI Japanese tutor as a tool, with customizable characters and scenarios, no companionship |
| Duolingo video call (Lily) | Duolingo Korea store 246k ratings | Also a fixed cartoon character, but you need a Max all-subjects subscription, and Lily goes the gloomy route with no affection system |
| 일본어는 이로이로 | Released 2026-01, 1,110 ratings on the Korea store, #24 in Education Top Grossing | The Japanese app growing fastest in Korea this year. It teaches JLPT vocabulary and grammar, not conversation |
| 제타 zeta | 70k ratings on the Korea store | The biggest AI character chat app. It doesn't teach anything |
| Big-tech companion products like Grok Companions | — | Users use it as the benchmark. Nobody offers Japanese lessons for Korean fans |

Source: public App Store Korea and Japan store pages and charts (11-18). Separately, Japan's Ministry of Foreign Affairs and the Japan Foundation are planning an app to learn Japanese by talking with popular anime characters through AI. It uses licensed IP and targets young overseas learners. They applied for ¥2027B in FY1.83 funding, with completion expected in FY2028 (Japan Times, 11-09). The idea has official backing, but launch is two years away.

A Japanese conversation app built for Korean anime fans, with one heroine and a progression system. So far, only 하나비 is in this space. Nobody knows about it: no press coverage, and its own accounts have only dozens of followers.

### What's still unclear

- Whether it can scale on its own. So far it has gotten seen mainly through the low bar of the Korean education charts and comments. The differentiation itself hasn't spread yet.

- Retention. Whether the design of 'free 5 minutes a day' brings people back every day or lets them leave after a chat is unknown. There is no data.

- How many people pay.

### If it were you

**B, worth a try.** The formula of 'one fixed character + a course + mobile-game-style progression and monetization' is worth trying in other 'subculture × language' combinations, and almost nobody is doing it. But 하나비 itself has only a dozen-odd ratings, the formula hasn't been proven to scale, and character art and voice are real costs.

**Who it suits:** Developers who are part of a subculture and know what this crowd likes in characters; people who can get decent character art and voice (drawing it themselves, having an artist friend, or willing to pay). Ideally someone who can build a bit of gamified system.

**Who it doesn't suit:** Anyone wanting a general AI speaking tutor (Speak, Duolingo and Amiko are already here, and big-tech companions can add courses at any time); anyone who plans to cobble together art and voice from free assets, because these users are buying the character, and no review praises the lessons; anyone who can't handle being criticized for 'raising a waifu inside a learning app.'

**How to do it:**

- Pick a different combination rather than head-to-head against 하나비 in Korea. The two most obvious are: anime fans in English-speaking markets learning Japanese (people are already asking for an English version in the reviews), and K-pop fans learning Korean (with idol trainees as the characters). Chinese anime fans learning Japanese also works, but payments and distribution are separate issues.

- Only one character, and it needs a backstory and setting. Affection levels, unlockable voice lines, side stories and memory are what make people come back every day, so they need to be in the first version.

- Courses don't need to be extensive, but they need a framing: 'what you learn today, she says to you today.' That links studying and chatting.

- Sell by the minute: a few free minutes a day, and a subscription-free monthly pass plus gem tiers, priced against the local mobile-game top-up tiers of the target market. Voice models charge by the minute, so billing should also be by the minute.

**How to acquire users:** Following 하나비's playbook, it costs almost nothing. Comment on the anime, Japanese-learning and idol short-video posts where the target audience gathers, using the same line. Have the character show up on current events within the fandom (competitions, new seasons, concerts) and teach a couple of words from that day. Make a vertical demo where the character moves and speaks. Use a separate short link for each channel and track them separately. Only consider paid acquisition once a channel's conversion is proven.

**When it has to be done by:** Get the first version out before the first half of 2027 (estimate). If you wait longer, two things can happen: big-tech companions (Grok, ChatGPT voice) add courses and local-language assistance; or 하나비 or Amiko release an English version. The Japanese government app is 2028 away, so don't wait for it.

**Signs that would mean you read it wrong:**

- By next January, if 하나비's Korea ratings are still in the dozens and it has dropped out of the top 200 in Education Top Grossing, that shows this combination can't scale even with the audience it's most suited to;

- If under one in ten of your free users are still coming back to chat on day 7, the character isn't keeping people, and 'raising a waifu' is just novelty;

- If Amiko or Duolingo launches a single-character plus affection-level feature and climbs the education charts in Korea or the US, that shows well-resourced players have entered, and the window is closing;

- Conversely, if 하나비 breaks into the top 50 of Korea's Education Top Grossing within two or three months, the formula works and you should speed up.

---

Data and sources: 하나비's App Store Korea page, reviews and in-app purchase list (11-14 to 11-18); Google Play page (11-18); official site hanabiapp.com; public pages for @hanabiappkr on Instagram, YouTube and X; search-engine snapshots of comments; 도도토's revenue from its Instagram posts and totodo.kr; competitor figures from public App Store Korea and Japan pages (11-18); Japan's Ministry of Foreign Affairs plan from Japan Times 2026-11-09. Anywhere marked 'estimate' or 'inferred' is an estimate or inference.

## Ads it's running

**Meta** has shown us 1 ads; 1 are still running, earliest from Nov 2026 15.

1. **Meta** · Video · Running for 6 days · Still running · [View original ↗](https://www.facebook.com/ads/library/?id=1808780240387388)
  일본어 공부 재미없으신가요? 이제는 미소녀 유이와 함께 즐거운 시간 보내며 일본어 공부해요!

These are the longest-running ads in this set. Long runs usually mean the creative is working. Data comes from the official Meta, Google and TikTok ad libraries and updates daily; open the official library to view the original creative.

[iOS store page ↗](https://apps.apple.com/app/id6828335335)

**Research notes: competitor table, acquisition evidence, full research report**

Competitors

| Name | Where | Launched | Scale | How it differs from this one |
|---|---|---|---|---|
| Amiko 아미코 | iOS | 2025-10 | iOS Korea store: 64 ratings; Japan store: 351 | Multi-character, 15-language scenario roleplay. 하나비 has one heroine, teaches only Japanese, targets Korean anime fans specifically, and has affection-based progression |
| 일본어 회화 도우미 아오이(DeepBrain) | iOS | About 2022 | 44 ratings in the Korean store | Tool-style AI Japanese tutor. Characters and scenarios are customizable, with no companionship or progression |
| Duolingo Video Call with Lily | Big-company features | 2024 | Duolingo Korea store 246k ratings | Needs a Max all-subjects subscription. Lily has no affection system or cute progression, and isn't aimed at anime fans |
| 스픽 Speak | iOS | 2017 | Korea store: 174k ratings, #2 in Education Top Grossing | Top Korean AI speaking app, focused on English |
| 헤이링 Heyring | iOS | 2024-07 | Korea store: 2,432 ratings, #6 in Education Top Grossing | AI phone-call foreign language learning (including Japanese), no character progression |
| 일본어는 이로이로 | iOS | 2026-01 | Korea store: 1,110 ratings and 7 days +166, #24 in Education Top Grossing | Teaches JLPT vocabulary and grammar, not conversation |
| 오레노 니홍고(도도토) | Web | 2024 | Publicly reports revenue of ₩50M to ₩100M | Real-person influencer-recorded otaku Japanese lessons. Same audience, but not AI conversation |
| Langaku 란가쿠 | iOS | 2022 | Japan store: 10,765 ratings; Korea store: 12 | Reading-based learning using Shueisha licensed manga, no conversation |
| 제타 zeta(Scatter Lab) | iOS | 2022 | 70,884 ratings in the Korean store | AI character chat that doesn't teach anything. No courses or corrections |
| Cotomo | iOS | 2024-02 | 7,852 ratings in the Japan store | A Japanese voice companion AI, not a learning product. No Korean-language support |
| Grok Companions / ChatGPT voice | Big-company features | 2025 | — | General companions or assistants with no Japanese courses for Korean users. Users compare 하나비 to 'a Grok companion that can teach Japanese' |
| Japan Ministry of Foreign Affairs / Japan Foundation anime-character AI Japanese app | Other | Planned for FY2028 | Applied for ¥1.83B in FY2027 | Uses licensed popular IP, targets young overseas learners, and still two years from release |

Where users come from

- **Verified** **Instagram native Reels** : @hanabiappkr posted a Reel teaching Japanese with 'praying to fox god Yui' from an LCK world championship (79 likes), and a Live2D VTuber-style demo (43 likes), with the uniform line 'link in profile #서브컬처 #하나비'

- **Verified** **Instagram comment-section link seeding** : the same line 'I practice with Yui in real-time conversation in the Japanese conversation app 하나비, link in profile' appeared under older viral Reels from a Japanese-learning influencer (도도토, DPP8Qj5k6Rm) and a SOOP streamer 쿠빈 (DQFsBVVEwQf) (search-engine snapshots)

- **Verified** **YouTube Shorts / X** : @hanabiappkr: 3 Shorts, top 1.4K views, 23 subscribers; X account with 62 followers

- **Verified** **Per-channel short links** : go.hanabiapp.com/web/play, /web/appstore, /x, tracking conversion by channel

- Inferred **Store search** : the title includes '일본어 회화'; a Google search for related terms returns the Play page first

- Inferred **Community or acquaintance-driven ratings** : 11-17 added +14 ratings in one day. Reviews cluster around 11-14/15, and the nicknames have anime flavor, but there are also specific bug reports

- Inferred **Paid acquisition** : not tracked in the metaads library, and the scale doesn't look like paid ads. TikTok not checked (TikHub 402)

## 하나비 - Japanese conversation with Yui (Hanabi): why it's getting traction

> iOS id6828335335 · Developer YongJin Kwag (곽용진, sole proprietor, Icheon, Gyeonggi-do; the developer name on Google Play is Idol Attention) · Released 2026-11-06 · Korea store · Education (secondary category Entertainment). Research date 2026-11-17/18. Data sources: appstore / googleplay / metaads library (agent-im server), App Store and Google Play pages, official site, public Instagram/YouTube/X pages, Jina/Brave search.

### Bottom line

It **isn't really 'taking off' yet** : it's been out for 12 days, with only 16 ratings on iOS and 31 installs on Play. It's getting attention because the bar for Korea's education charts is low. 11-17 gained 14 ratings in one day, which pushed it to #51 in Education Free and #126–169 in Education Top Grossing. What's worth watching is the playbook: a solo studio started in June 2026 that combines 'subculture companionship (waifu character, Live2D, affection, unlockable voice, memory)' with 'a course running from kana to travel conversation' around one fixed heroine. It monetizes through Korean mobile-game gem tiers plus a '30-day no-renewal pass,' and recruits users by posting uniform lines in comments on anime/Japanese Reels and riding current-event short videos (LCK world championship). Early reactions ('내 아내임', '하나비남친', '미소녀다이스키') suggest the positioning hits Korea's 서브컬처 (subculture) crowd, but whether it can scale is still unclear.

### 1. Data: how big it is now

| Metric | Value | Source |
|---|---|---|
| iOS ratings | 16 in total, average rating 4.875 (11-14 at 0 → 11-16 at 2 → 11-17 at 16 → still 16 at 11-18) | appstore library app_day(kr); iTunes lookup |
| iOS Korea store, Education Top Free | 11-16 #53 → 11-17 #51 → 11-18 #56 | appstore library, chart_rank |
| iOS Korea store, Education Top Grossing | 11-16 #126 → 11-17 #159 → 11-18 #169 | appstore library, chart_rank |
| Countries charted | Korea only | appstore library |
| Google Play | Installs 27(11-16) → 31(11-17, 11-18), pricing tier 10+; best Korea Education Top Free rank #45 | googleplay library app_day / app_profile |
| Release cadence | 1.0.1(11-08) → 1.0.2 → 1.0.3(11-14/16) → 1.0.4(11-17), new version every 10 days, 4 releases | App Store version history, Play page |
| Package size | 337 MB (Live2D + voice assets), iOS 15+, rated 12+ (notes "occasional sexual suggestive themes") | App Store page |

How to read these numbers:

- Education Top Grossing #126 with only 2 ratings, which suggests **people are already paying** (even if just a few); but the bar to chart in Korea's Education Top Grossing past the hundredth spot is low, so this doesn't tell us the revenue scale.

- 11-17 gained 14 ratings in one day. The App Store shows 10 written reviews, all timestamped "2–3 days ago" when scraped, meaning they're concentrated between 11-14 and 11-15. This is a burst: it could be a push or a piece of content bringing in a community, or some of it could be friends or people who were directed there. I can't tell which yet.

- It doesn't appear in the metaads library (the library only tracks listed advertisers, so not finding it doesn't mean it isn't advertising). TikTok couldn't be checked (old TikHub account out of balance, returned 402).

### 2. What it is and how it makes money

**Product** (official site hanabiapp.com, store description):

- Only one heroine, "유이 (Yui)", a fox god at "Hanabi Shrine." The product explicitly says "캐릭터를 고르지 않습니다" (you don't choose a character). Live2D illustration; tap or touch her and her expression changes.

- **Live practice** : speak or type Japanese to Yui by voice or text, and she answers by voice; her lines come with kana readings, Korean translation, and Korean-script pronunciation; if you get stuck, you can pick suggested replies. After you speak, it rewrites your sentence into natural Japanese and gives a score report on expressiveness and completeness.

- **Lessons** : hiragana stroke-order tracing → graded vocabulary → scenario-based conversation stages (greetings, convenience store, café, asking directions). Expressions you learn that day show up in that day's conversations with Yui.

- **Companion system** : affection levels that unlock new lines and voices (1.0.4 update: after an affection level-up, it shows on the home screen directly, and new voices are tagged NEW), Yui's side stories. While you hold Max or the unlimited pass, Yui remembers what you talked about. There are also daily missions and a streak counter.

- Still labeled "오픈 베타" (open beta).

**Pricing** (App Store Korea store, in-app purchase list):

- Free: 5 minutes of live practice with Yui per day.

- 30-day passes, **no auto-renewal** : Plus ₩4,900 (10 min/day), Pro ₩9,900 (20 min), Max ₩19,900 (40 min + memory), Unlimited ₩39,900.

- Gems (used to add conversation time): 150 for ₩1,200 / 830 for ₩6,000 / 1,800 for ₩12,000 / 4,000 for ₩25,000 / 10,700 for ₩59,000 / 23,800 for ₩119,000.

This price set follows **the standard top-up tiers of Korean mobile games** (1,200 / 6,000 / 12,000 / 25,000 / 59,000 / 119,000), plus a "monthly card"-style non-renewing pass, rather than the weekly or annual subscriptions common in education apps. Pricing by minute also makes sense: real-time voice and LLM calls cost scale with duration, so selling by time locks in margin, and the top tier at ₩119,000 is for heavy users.

### 3. Competitors: who else does the same thing

| Competitors | Where | Launched | Scale (source) | How it differs from Hanabi |
|---|---|---|---|---|
| Amiko 아미코:AI 일본어·영어회화 채팅 연습 | iOS/Android | 2025-10-22 | iOS Korea store 64 ratings, Japan store 351 ratings (appstore library snapshot) | The closest match: also "practice Japanese with an AI character," with a Korean landing page. But it's multi-character (café clerk, "husband at home," transfer student...) and multi-language (15 languages), a scenario roleplay product. Hanabi has one heroine, teaches only Japanese, and targets Korean anime fans. |
| 일본어 회화 도우미 아오이(DeepBrain LLC) | iOS | around 2022 | Korea store 44 ratings (appstore library snapshot); Naver has annual-subscription reviews from paying users | A tool-style AI Japanese tutor where you can customize characters and scenarios, with no companionship or progression system |
| Duolingo Video Call with Lily | Big-company features | From 2024, iOS supports Japanese and Korean | Duolingo Korea store 246k ratings, 7-day +2,728 (appstore library) | Also "video call with a fixed cartoon character," but you need the Max subscription, and Lily is a gloomy character with no affection system. In Korea Duolingo is a catch-all entry point, not aimed at anime fans. |
| 스픽 Speak | iOS | 2017 | Korea store 174k ratings, Education Top Grossing #2 (appstore library) | The leader in Korean AI speaking, but focused on English; shows Korean users will pay for AI speaking practice |
| 헤이링 Heyring(HYPERNOVA) | iOS | 2024-07 | Korea store 2,432 ratings, Education Top Grossing #6 (appstore library) | AI phone-call language practice (including Japanese), taking a "phone English" substitute approach, with no character progression |
| 일본어는 이로이로(Irolabs) | iOS | 2026-01 | Korea store 1,110 ratings, 7-day +166, Education Top Grossing #24 (appstore library) | The Japanese app rising fastest in Korea this year, but it does JLPT vocabulary, kanji and grammar, not conversation |
| 오레노 니홍고(도도토) | Web courses + YouTube/Instagram | 2024 | Publicly says revenue passed 50M won (Instagram post); website says it passed 100M won (totodo.kr) | "Japanese class just for otaku," taught by a real influencer on camera. It proves Korean anime fans will pay for Japanese, and its audience fully overlaps with Hanabi's. Hanabi has also left comments under its Reels. |
| Langaku 란가쿠 (MANTRA, licensed from Shueisha) | iOS/Android | 2022 | Japan store 10,765 ratings (mainly users in Japan learning English); Korea store 12; 2026-10 finds creators to promote it in Korea | Learn Japanese with licensed manga: read comics, look up lines, no conversation |
| 제타 zeta(Scatter Lab) | iOS/Android | 2022 | Korea store 70,884 ratings, Japan store 216k (appstore library snapshot) | Korea's biggest AI character chat app, showing young Koreans love chatting with AI characters; but it teaches nothing and doesn't sell itself as learning |
| Cotomo(Starley) | iOS | 2024-02 | Japan store 7,852 ratings (appstore library) | A voice companion AI in Japan; some people on the Korean DC Japanese-learning board use it to practice conversation, but it isn't a learning product and has no Korean-language support |
| Grok Companions (Ani etc.) / ChatGPT voice | Big-company features | 2025 | — | One review says: "I want to unlock outfits by affection like Grok companion," showing users treat Hanabi as "Grok Ani that teaches Japanese." Big-tech general companions don't offer Korean-supported Japanese lessons. |
| Japan Ministry of Foreign Affairs / Japan Foundation anime-character AI Japanese app | Government project | Planned completion FY2028 | 2026-11-09 reported by Japan Times, FY2027 application 1.83B yen | Almost the same idea (learn Japanese by talking to a popular anime character with AI), but aimed at overseas young people, using official IP, and two years out. It also suggests the direction is officially recognized as a real pain point. |

Takeaway: "AI Japanese conversation" is not short on products in Korea, and "AI character chat" has big players like Zeta, but I've only found Hanabi doing **a Japanese conversation app built for Korean anime fans, with one heroine and progression** . The nearest, Amiko, goes multi-character and multi-language.

### 4. Growth levers: where users come from

| Channel | Evidence | Confirmed? |
|---|---|---|
| Posts Reels from its own Instagram account @hanabiappkr | Uses one standard caption: "일본어 회화 앱 하나비 유이와 실시간 대화로 연습합니다 링크는 프로필에 #서브컬처 #하나비" (Practice Japanese by talking live with Yui from the Hanabi app, link in profile). It piggybacked on the LCK World Championship final (11-15 Gen.G vs BLG), posting "Praying to fox god Yui for an LCK title" and teaching "神頼み / 必勝祈願 / がんばれ" along the way: 79 likes, 2 comments (5 days ago when scraped). About 11-16 it posted a Live2D face-capture VTuber-style Reel with 43 likes; the pinned comment reads "구라 안까고 이번 기말 1등급 맞아옴" (no lie, got a grade 1 on this final exam). | Confirmed |
| Spamming links in the comments of other people's viral Reels | Search engines indexed the same line "일본어 회화 앱 하나비 유이와 실시간 대화로 연습합니다 링크는 프로필에 #서브컬처 #하나비 · hanabiappkr" appearing under the Reel of Japanese influencer 도도토 (DPP8Qj5k6Rm, "최고의 일본어 AI" / "best Japanese AI," 59 weeks ago) and under the Reel of SOOP streamer 쿠빈 (DQFsBVVEwQf, "애니로 일본어 공부했더니" / "studied Japanese with anime," 56 weeks ago); when recalled it also showed up under unrelated accounts like 잡지식 and 스시자로쨩. | Confirmed (two cases with search snapshots; the page only shows top comments, so the original comments aren't visible) |
| YouTube Shorts @hanabiappkr | 3 Shorts, views 473 / 929 / 1.4K, 23 subscribers | Confirmed, but very small volume. |
| X @hanabiappkr | 2026-11 registrations, 4 posts, 62 followers | Confirmed, but very small volume. |
| Short links tracked per channel | go.hanabiapp.com/web/play, /web/appstore, /x, etc., tracked separately per channel | confirmed (it's measuring conversion across channels) |
| Store search | the title stuffs in the keyword "일본어 회화"; search "하나비 일본어 회화 앱" on Google and the Play page is the first result | Inferred |
| Paid acquisition | Not in the metaads library (untracked); the store numbers don't look like large-scale paid acquisition either | Not found |
| rating surge | 11-17 gained +14 ratings in one day; the text reviews cluster around 11-14/15, with very uniform nicknames carrying a 2D/anime flavor (하나비남친, 미소녀다이스키, 바퀴벝레), but there's also concrete feedback (advanced mode speaks too fast, needs iPad support, foldable screen layout breaks) | Possibly community or acquaintance-driven prompting; not verified |

### 5. Why it's catching on (at its current small scale)

1. **The positioning is the users' own words** : the store copy says "want to watch anime without subtitles and understand the songs you love, but get stuck before the vocab list", "deleted several vocab apps", "practicing conversation with real people is stressful". Reviews come back with "씹덕인데……실제로 대화하는 느낌", "유이짱 귀엽고", "내 아내임". The motive and the experience match, so the first users are willing to leave good reviews.

2. **The audience has proven it will pay** : 도도토's "오레노 니홍고" publicly reports 50M to 100M won in revenue, 제타 has 70k ratings in Korea, and Speak and 헤이링 both sit in the top ten of the education Top Grossing chart. 하나비 sits at the intersection of three demands already validated: "2D culture × Japanese × AI speaking practice".

3. **It puts companion-game retention mechanics inside a learning product** : affection levels, unlockable voice lines and side stories, memory at the Max tier, daily tasks and streak counts. Users raise the character like a companion rather than using a tool, which is its core difference from tool-style products like Amiko and 아오이.

4. **Gamified pricing** : a non-renewing monthly pass plus mobile-game gem tiers matches how Korean anime fans already pay (like gacha games), and it sidesteps users' dislike of auto-renewing subscriptions; selling by the minute keeps the cost of voice AI under control. It hit the Korea education Top Grossing chart (#126) on day 10 after launch, which shows people are buying.

5. **Cheap, targeted distribution** : seeding uniform pitches in the comment sections of anime and Japanese-learning Reels where the audience gathers, plus short videos riding current events (LCK World Championship) and a VTuber-style Live2D demo. Very little money spent, well-aimed, still small in scale.

6. **Fast iteration** : 4 versions in 10 days, reworked around early feedback (reconnection, affection hints, keyboard, Google sign-in).

### Six: differentiation, and whether it's the reason it took off

The differentiation = **"one fixed anime heroine + a systematic Japanese course + mobile-game-style progression and monetization"** , for Koreans learning Japanese only. In comparison:

- Against AI Japanese conversation tools (아오이, Amiko, Kotogoo, Speak-style products), it sells "a relationship with a character", not "practice reps".

- Against AI character chat (제타, Grok companion, Cotomo), it has a course, corrections, and the legitimate frame of "studying Japanese" (users can justify it to themselves and share it publicly).

- Against Duolingo's Lily, its character is "moe + affection levels", and you don't need to buy a full-subject subscription.

This differentiation **really is the source of early word of mouth** : the reviews mostly praise "유이 is cute", not the course. But the scale it has brought so far is only a dozen-odd ratings and a few dozen installs, nowhere near proof it can support a business. So far it has been noticed more through the low bar of Korea's education chart and comment-section seeding than through its differentiation spreading on its own.

### VII. Visibility and copycats

- **Visibility: few people know about it yet.** No media coverage, no public revenue, no mentions in the x-radar corpus on X; its own accounts have only a few dozen followers each.

- **Copycats: none.** Similar concepts exist, like Amiko (a year earlier, multi-character, multi-language) and Japan's government-planned anime-character app (FY2028), but neither is copying it.

### 8. What others can learn from it

- The formula "**single character + course + progression** " can be copied to other "subculture × language" pairs: anime fans in Europe and the US learning Japanese (commenters are already asking for an English version), 2D culture fans in China learning Japanese, Japanese people learning Korean (K-pop audience), fans learning Korean to follow idols.

- **Sell study time through mobile-game tiers** : a non-renewing monthly pass + gems, billed by the minute, so voice AI costs and revenue scale together.

- **Seed pitches under viral content where the audience already gathers** , plus character-driven current-events shorts (folding news events into "today's Japanese"). Cost is nearly zero, which suits cold start.

- **Per-channel short links** to track conversion, so even a small budget tells you which channel works.

### 9. What I'm still unsure about

- Whether the 11-17 14 ratings came in organically, from community posts, or from acquaintances or prompting (the reviews are uniform in style but also contain genuine usage feedback).

- Real download counts and paying users (the Top Grossing #126 bar is low, so revenue may be only tens of thousands of won per day; unverified).

- The developer 곽용진 shares a name with the CEO of 이르테크 (IIR TECH, a Daejeon AI language education company that makes AI writing and speaking correction tools and the AI teacher for 세종학당); but 하나비 is an individual business in 利川 (business ID 473-03-04113), while 이르테크 is a corporation (884-86-***** ). Whether they are the same person is **unverified** . If so, it has an NLP and language-education background behind it. The GitHub organization IDOL-ATTENTION was created in 2026-06-16, suggesting roughly 5 months of development.

- Whether it runs ads on Instagram or TikTok (not tracked in the metaads library; TikTok not checked).

- Daily retention: whether the design of 5 minutes of free use per day keeps people coming back; no data yet.

### Sources

- App Store Korea page, reviews page, in-app purchase list: [apps.apple.com/kr/app/id6828335335](https://apps.apple.com/kr/app/id6828335335) (2026-11-18 scraped)

- Google Play:[play.google.com/store/apps/details?id=com.idolattention.hana…](https://play.google.com/store/apps/details?id=com.idolattention.hanabi)

- Website: [hanabiapp.com/](https://hanabiapp.com/)

- Instagram: [www.instagram.com/hanabiappkr/](https://www.instagram.com/hanabiappkr/); LCK Reel [www.instagram.com/reel/Dfo4dlUI92c/](https://www.instagram.com/reel/Dfo4dlUI92c/); Live2D Reel [www.instagram.com/reel/DfzpJHLIBid/](https://www.instagram.com/reel/DfzpJHLIBid/); comment-section seeding links [www.instagram.com/reel/DPP8Qj5k6Rm/](https://www.instagram.com/reel/DPP8Qj5k6Rm/), [www.instagram.com/reel/DQFsBVVEwQf/](https://www.instagram.com/reel/DQFsBVVEwQf/) (Jina search snapshots)

- YouTube:[www.youtube.com/@hanabiappkr](https://www.youtube.com/@hanabiappkr) ;X:[x.com/hanabiappkr](https://x.com/hanabiappkr)

- 도도토 revenue disclosure: [www.instagram.com/p/DBX1Capzr7R/](https://www.instagram.com/p/DBX1Capzr7R/), [totodo.kr/](https://totodo.kr/)

- Amiko:[amikoapp.com/ko/learn-japanese](https://amikoapp.com/ko/learn-japanese)

- Duolingo Video Call:[blog.duolingo.com/video-call/](https://blog.duolingo.com/video-call/)

- Japanese government anime AI Japanese app: [www.japantimes.co.jp/news/2026/11/09/japan/media/mofa-japane…](https://www.japantimes.co.jp/news/2026/11/09/japan/media/mofa-japanese-app/)

- 이르테크/곽용진: [www.veritas-a.com/news/articleView.html?idxno=552209](https://www.veritas-a.com/news/articleView.html?idxno=552209), [app.rndcircle.io/company/8172d783-e857-588c-a490-41b781d3976…](https://app.rndcircle.io/company/8172d783-e857-588c-a490-41b781d3976b)

- Databases: appstore (app_day, chart_rank, app_profile, app_snapshot), googleplay (app_profile, app_day), metaads (app_link); 2026-11-18 queries

This article was written by AI from public data and the data we collect daily. Numbers are sourced, and inferences are labeled.
