ゆめ · a dream shaped by wordsYume
Let's play! Pick a worldSay it in Japanese (or English) and the world changes: a whale leaps, night falls, a magic door opens. Your words are the magic.
version 0.1 · early previewChoose a world
One minute of real play
A whale that leaps when you tell it to, a quest, a magic door to the sky islands, a dragon, and Hina, who remembers you.
Recorded live on Orbis. Music, the Honey Bird and the sky were made with fal; the voice is Yume's own narrator.
A game, not a lesson
Quests
The world asks for help: the whale is shy! Say the right words and win a gold sticker.
A magic door
Say ドアを あける and step into a whole new world: a toy room, a candy town, the moon.
A sticker book
Everything your words make becomes a sticker, with a photo of the world you made it in.
Your dream movie
When the dream ends, it turns into a short film of your best moments, ready to share.
Hina remembers
Your friend Hina remembers your adventures: “Last time you made a whale jump!”
Play together
Two players, one world: one learns Japanese, one learns English, taking turns.
Every word becomes a sticker
55 to find, and anything new you dream up gets drawn just for you.
Pre-rendered video can't do this
The learner's next sentence is unpredictable, every single turn. Yume keeps one running world alive and steers it in real time as each answer lands — that's not a visual flourish, it's the entire teaching mechanic.
01Correct answers move the world
A right sentence steers the live Orbis scene within seconds — the same running world, not a new clip.
02Wrong answers change nothing
If the meaning doesn't land, the world holds still. The video itself is the feedback.
03State persists, turn after turn
Every correct answer builds on the last — a cat, then a dog beside it, then rain — one continuous scene shaped entirely by language.
The stack
Next.js 16
App shell, UI, and API routes
Orbis · Reactor
Live, steerable video world over WebRTC
GPT-OSS 120B
Grades answers, narrates, writes quests and speaks as Hina, on Cerebras and Groq, with a fal backup
Fish Audio
A Japanese and an English voice for the narrator and Hina
fal.ai
Drew every sticker, background and mascot here, and draws new stickers while you play