CALL ME SENSEI — EARLY PROTOTYPE / ONE-DAY HACKATHON
Early prototype work for Call Me Sensei, the voice-first language-learning game — built with the team in a one-day hackathon.
PROJECT VIEW
Give her a gift, learn the word, say it out loud. Every layer of the prototype is AI — the character videos, the live persona, and the real-time voice.
GAME DESIGN / VOICE-DRIVEN LOOP
The loop is gift → tap → talk: pick a gift, she turns it into a Japanese vocabulary lesson, and the talk phase asks you to actually speak. Affinity builds with streaks and drives her dual personality — keep feeding her Japanese and the demure mask cracks into Yandere Mode, where she visibly feeds on your learning energy.
LUMA DREAM MACHINE + KLING AI / IMAGE-TO-VIDEO
Kotodama is pure video: every state the Phaser canvas shows — idle, talking, receiving gifts, powering up, transforming, feeding — is an MP4 clip. The hard part of AI video is keeping one continuous character across every clip and across two personalities; that consistency was the whole job.
MY ROLE / SCOPEMy contribution: the unified character concept and all animation clips — hand-recorded reference performances first, then image-to-video generation with Luma Dream Machine and Kling AI.
ARCHIVE NOTEAI disclosure: all character animation clips were generated with Luma Dream Machine and Kling AI from hand-recorded reference performances. Workflow-test material from those image-to-video sessions may be added here later.
LLM PERSONA + EMOTION-MAPPED TTS
Her lines are written live by an LLM persona that escalates through affinity stages, then spoken through real-time TTS whose voice direction is remapped per emotion — sweet, sweet-encouraging, and hungry levels that surface as the player keeps speaking Japanese. The player's own voice comes in through browser speech recognition with a language detector: literally speaking Japanese is what changes her.
const MODEL = 'gpt-4o-mini-tts'
const DEFAULT_VOICE = 'coral'
const EMOTION_PRESETS = {
// ── Sweet Mode (default / English input) ──
// The mask she wears. Cute, polite, slightly too perfect.
sweet: {
instructions:
'You are a sweet, soft-spoken young Japanese woman. Speak gently and politely with a '
+ 'cute, slightly breathy quality — like an anime girl being demure. Your tone is warm '
+ 'and inviting but just a little too sweet, as if you are carefully performing sweetness. '
+ 'Occasionally let a tiny sigh or soft breath slip through between sentences.',
speed: 1.0,
},
sweet_encouraging: {
instructions:
'You are a sweet young Japanese woman praising someone you adore. Sound genuinely '
+ 'delighted and proud — your voice lifts with excitement. '
+ 'But underneath the celebration there is a subtle hunger — you want them to do it again.',
speed: 1.05,
},
// ── Hungry Mode (Japanese input detected) — intensity scales with affinity ──
// The mask cracks. Each level is more exposed.
hungry_1: {
instructions:
'Your voice catches slightly — a small gasp of surprised delight before '
+ 'you compose yourself. Still polite, but your breath comes a little faster.',
speed: 1.0,
},
hungry_2: {
instructions:
'Your voice is warmer now, leaning in — you speak a little faster because you cannot '
+ 'help yourself. Your politeness is still intact but thinner — the real you is peeking through.',
speed: 1.05,
},
hungry_3: {
instructions:
'You are barely holding herself together. Your voice trembles with need that you are '
+ 'trying to disguise as enthusiasm. Breathy, urgent.',
speed: 1.08,
},
}KEEP EXPLORING