← Home / Culture

Impressive·Culture· Aug 5, 2026 · 7 min

The Idol Who Can't Code Just Built an Entire Livestream System With AI

A former Hello! Project idol ran a 10-hour, fan-powered livestream on a platform she built herself — and says she never read a single line of the code.

The Idol Who Can't Code Just Built an Entire Livestream System With AI

What's going on

Miyamoto Karin joined the Hello! Project idol pipeline at age nine in 2008, eventually became a founding member and de facto ace of the group Juice=Juice, and graduated from it in 2020 to go solo. Among fans she has a reputation for being unusually serious and studious on stage — which makes her about the last person anyone expected to start dropping words like "API" into casual conversation.

For the release week of her fifth single, "HANAKIN" (a play on 花金, Japanese slang for the giddy feeling of a Friday evening), she ran a 10-hour YouTube livestream built around a fan challenge: recreate her music video using only things lying around the house. The twist was a built-in incentive system — every time a viewer bought her CD and posted about it online, a gauge on screen climbed, and each new threshold unlocked another piece of the production, from costumes to makeup to editing software.

The stream itself made headlines, but what really got people talking came after: Miyamoto, who has no programming background and says she never once opened the actual code, published a full technical breakdown of everything running behind the scenes — a serverless backend, a live overlay for her broadcast software, a phone-based control panel, and an AI system scoring submitted videos, all built by describing what she wanted, in Japanese, to an AI coding assistant.

Comments

@karin__miyamoto — Official

The story of an idol who can’t code, building an entire 10-hour livestream system together with AI.

I also wrote up the technical side of it.

Please go take a look!

コードが書けないアイドルが、AIと一緒に10時間配信のシステムを全部作った話。 技術面でまとめたものも書きました。 是非見てみてね!
You should post this on Qiita too! (Qiita is a Japanese platform where engineers share technical write-ups)
Qiitaにも投稿しよう!
Karin-chan is amazing.
佳林ちゃん凄すぎ〜。
Is she seriously the first idol in history to write a technical blog…?
技術ブログ書くアイドル、史上初じゃない…?
Doing it all by herself is amazing! Actually it’s kind of insane (lol)
1人でやっちゃうところが凄い!というかおかしい(笑)
Never thought I’d see the day when a former Hello! Project member was out here talking about AI.
元ハロメンがAIのことを 発信する時代がくるとは。
Wow! I can’t count how many times Miyamoto-san’s disciplined, almost monk-like dedication and logical way of talking has surprised me.
すご!宮本さんの求道的な姿勢とロジカルな語りに何度驚かされてきたことか
This is too funny, first time I’ve ever seen the word “CORS” show up in an idol’s blog.
面白すぎる、アイドルのブログでCORSなんて単語出てくるん初めて見た
As someone who’s been watching her from way back, seeing Miyamoto Karin handle Cloudflare like a pro and turn her official blog into a tech blog is genuinely nothing but hope — I can’t watch it without tearing up.
昔から見てる身としては宮本佳林さんがCloudflareを使いこなし公式ブログが技術ブログと化してるの本当に「希望」しか無く、涙無しには見れない
It’s funny that right when people are up in arms saying “don’t you dare use generative AI in a Juice=Juice music video,” former Juice=Juice super-ace Miyamoto Karin goes and builds her whole content pipeline with generative AI.
Juice=JuiceのMVに生成AI使うの許さんって燃えてる時に元Juice=Juiceのスーパーエース宮本佳林が生成AIをフル活用してコンテンツを作る流れ面白い。
This is insane… Juice=Juice, which is having a huge moment right now, practically felt like it was built around Miyamoto Karin — a monk-like super-ace idol — and then she suddenly announced she was graduating to “do DIY and make things,” which sent fans into a stir. But now she’s literally building technology and making things happen for real, and that’s just cool.
すごすぎる…。今大バズり中のJuice=Juiceは宮本佳林ちゃんのために作られたくらいのスーパーエース・アイドル求道者なのに突然「DIYとかもの作りする」と言って卒業してオタクをざわつかせていたのですが、テクノロジーを作ってしっかりもの作りを実現しててかっこいい
Ohhh, I see — in the middle of this whole “feeding your MV into generative AI is bad” sentiment, she’s the one actually building the app that feeds her own MV into a generative AI (LLM). That’s hilarious.
ああーーこれ、「MVを生成AIに食わせるの良くない」って風潮がある中で「自分のMVを生成AI(LLM)に食わせるアプリを作る」のを本人がやってるのか。おもろ

This is genuinely impressive — and it’s not just “she generated some code with Claude Code” 🤯

In ten days she put together Cloudflare Workers + KV, an on-stream HUD for OBS, a phone-based admin dashboard, a system for pulling post counts from X, and even a local app that scores video with a Vision API.

Her planning ability and judgment calls are what’s actually incredible here.

これは本当にすごい、単に「Claude Codeでコードを生成した」話じゃない🤯 10日間で、Cloudflare Workers+KV、OBS用HUD、スマホ管理画面、X投稿数の取得、動画をVision APIで採点するローカルアプリまで組んでいる。 この方の本人の企画力と判断力がそもそも凄すぎる。

Still haven’t fully read it, but the more I read into the details the more shocked I get.

Honestly it’s shocking enough that the word “API” is coming out of Karin-chan’s mouth at all — an API is basically the interface that lets a web service talk in requests and responses instead of going through a browser (that’s what my own YouTube search does too).

まだ十分に読めてないけど詳細を読めば読むほど改めて衝撃。 そもそもAPIというワードが佳林ちゃんから出るのが衝撃なんだけど、APIとはウェブサービスについてブラウザ経由でなくリクエスト/レスポンスでやり取りするためのインターフェースのこと(私のYouTube検索もそれ)。

Wow… she’s genuinely, seriously talented…

I handed all the implementation over to Claude Code and never read a single line of the code myself, so I can’t really explain “why it’s written this way.” Still, what’s running where, why I chose to put things there, where I got stuck on the day — those are the calls I made myself, so I’m writing down what I can.

すごい…ほんとマジで優秀な方なんだなぁ… >実装はすべてClaude Codeに投げていて、コードは1行も読んでいないので、「なぜこう書くのか」を説明できるわけではないんです。 >それでも、何がどこで動いていて、なぜその置き場所を選んだのか、当日どこでつまずいたのか─ >そこは自分で判断してきたので、分かる範囲で書き残しておきます。

I’m an engineer myself! Even though the post barely uses any technical jargon, you can tell she genuinely understands what she built. It almost feels like she’s deliberately holding back the terminology because it’d be overkill here.

It’s also great that she had a clear requirement of her own — not wanting the stream to stop.

I think you can build pretty much anything, as long as it’s not handling other people’s money or data.

私エンジニアです! 記事に専門用語が出てきてないのに、本当に作ったのが分かるくらい理解されてる。用語はここでは過剰だからわざと控えているという配慮すら感じる。 配信中に止めたくないって言う自分の持つ要件が明確なのも素晴らしい。 人のお金やデータを預かるもの以外は何でも出来ると思う

Honestly this feels like it could pass as a science-major graduation thesis 😂

Then again, that also means someone with zero technical knowledge can pull this off with AI, which got me curious about how universities handle graduation theses these days — turns out it’s a whole mess of “partial use allowed” rules and so on…

Though if you frame it as research about using generative AI, it’d probably fly.

普通に理系の大学での卒論として通っちゃう気がする内容😂 逆に考えると、技術的知識がゼロの状態でもAI使えばできちゃうって事なので、今の大学の現場の卒論とかってどんな感じなのか調べてみたら部分的に使用可とか色々大変なのねぇ・・・。 生成AIを使うことに関する研究としてなら多分いける。
Truth.
真理だ

This is so insane my jaw actually came unhinged.

Knowledge is basically free now, and “attention to detail” is turning into the only thing that sets you apart —

guess that era has officially begun.

凄すぎてアゴ外れた 知識はほぼ無料になって 「こだわり」だけが差別化要素になっていく時代 始まったんだな

The story of an idol building an entire livestream system with AI.

“The Technical Architecture of the HANAKIN Livestream System.”

It’s amazing, I’d definitely recommend reading it.

アイドルがAIと配信のシステムを全部作った話 HANAKIN配信システムの技術構成 スゴイので読むのお勧め。

Amazing, seriously amazing.

I feel this myself from using AI a lot — fundamentally it’s the strongest tool there is for amplifying whatever you already have.

In other words:

  • If you’re already someone with something to offer, AI makes you unstoppable
  • If you’re currently nobody in particular, using AI still leaves you nobody in particular

That’s basically the story here. It’s a cruel era.

すごい、すごすぎる。 私も AI を触っていて実感しますが、本質的には「もともと持っているものを拡張するのに最強のツール」なんですよね。 つまり ・現時点で何者かである人がAIを使うと無敵 ・今の時点で何者でもない人がAIを使っても何者でもないよね というお話。 残酷な時代だ。

Well, at this point anyone can build basically anything, so it’s going to come down to how well you can multiply an LLM against whatever you’re already good at, and supercharge that strength.

In other words, it’s not just about the LLM by itself — the value of actually being a domain expert goes way up too, and combining the two makes you even stronger.

Honestly, not having that expertise is going to be the rough part now.

まあもうこの様に誰でも何かを作っていけるので、如何に自分の得意領域にLLMを乗算させて、その得意領域をガンガン強化させていくかってなっていくでしょう。 つまりLLM単体ではなくドメインエキスパートになっていく価値もめちゃくちゃ上がるし、掛け合わせるとさらに強い。 むしろないと辛い。
It’s genuinely amazing that AI has made it possible for literally anyone to turn their ideas into something real. It made me realize that whether you can code matters less than being able to concretely think through what you want to build and how you want it to behave, and put that into words. Going from the initial concept, to figuring out what features you need, to prioritizing them, all the way to actually shipping something that got used live for a 10-hour stream — that’s what’s really impressive.
AIがあることで、本当に誰でも、自分のアイデアを形にできる時代になっているのが本当にすごい。コードが書けるかよりも、何を作りたいのか、どう動いてほしいのかを具体的に考えて言葉にできることが一番重要になってきているんだなと思った。企画から必要な機能を考えて、優先順位をつけて、実際に10時間配信で使われるところまで作り切っているのがすごすぎる。
I even went and read the technical blog post! ✨ I was genuinely moved by the idea — and the follow-through — of turning AI into an ally precisely because she can’t code, and using that to pull off a 10-hour stream. Knowing what was happening behind the scenes only deepened my feelings for HANAKIN! Looking forward to whatever she takes on next 😊💜
技術ブログまで読ませてもらいました!✨ 「コードが書けない」からこそAIを味方にして、10時間配信を形にした発想と行動力に感動しました。配信の裏側を知ることで、さらにHANAKINへの想いが深まりました!これからの挑戦も楽しみにしています😊💜

My take

Turns out singing and dancing was never going to be the ceiling — apparently building your own tech stack is just part of the job description now. Japanese idols really know no bounds.
Source & translation note

Comments loosely translated for tone. Reactions collected from Togetter and X threads.

View original thread →