An AI-powered virtual idol platform that enables autonomous livestreaming and real-time fan interactions through natural conversations, expressive voices, and animated avatars.
Virtual idol content (VTuber) had strong audience demand in Japan — but every live session required a human operator behind the character at all times. One operator, one stream. Scaling to multiple simultaneous characters, or running extended broadcast hours, was not viable with that constraint.
The client's vision was an AI that could run its own sessions — reading viewer comments, generating contextually appropriate responses, and delivering them through voice and character animation — live, in real time, without a person managing each interaction. It also needed the full engagement mechanics that Japanese entertainment app audiences expected: gifting, fan rankings, direct messaging.
Building this well — not just as a proof of concept but as a product people would return to daily — took 24 months of continuous development.
AMCOLAB delivered the full platform end-to-end: from SRS through app store launch and ongoing operations. Five engineers have supported the product continuously, shipping new features and maintaining the live service.
AI characters broadcast live sessions autonomously. Viewer comments are read, processed through an LLM pipeline, and delivered as voiced character animation — within a latency window that feels natural in a live context. Characters adapt their content to what viewers request: gaming sessions, karaoke, conversation, or themed broadcasts.
Responses are delivered as emotionally modulated speech, synchronized to character animation in real time. Tone, pace, and expression vary with context — the AI idol sounds different when it is excited, empathetic, or playful.
Characters are rendered using Three.js with real-time animation sync. Lip movements, expressions, and gestures are driven by the voice output and conversation state. The animation layer is modular — new expressions and characters can be added without rebuilding the core system.
Viewers message AI idols directly in 1:1 chat or group chat during streams. Virtual gifts accumulate into a fan support ranking visible to all viewers — a mechanic that drives repeat session attendance and community engagement.
A schedule view shows upcoming sessions by character, theme, and time slot. Users set reminders for sessions they want to attend. The timetable drives daily active use beyond discovery-based browsing.
App publicly released on iOS and Google Play for the Japan market — actively operated
AI characters run live sessions autonomously — multiple characters can stream simultaneously with no proportional increase in operator headcount
Viewer engagement supported through layered interaction: live chat, gifts, fan rankings, 1:1 messaging
24-month continuous development cycle — the product matured significantly beyond its initial scope
Engineering team maintained stable context across the full engagement — no ramp-up cost from team changes
LLM orchestration, TTS, and real-time character systems built from scratch for production use, not adapted from demos
consumer-facing design for a Japanese entertainment audience; engagement mechanics designed for retention, not just functionality
ongoing feature development, operational support, and scaling as the product grew
SRS through App Store and Google Play submission support
real-time streaming platform for consumer audiences
AI applied to a consumer-facing two-sided platform
large-scale AI system built for sustained production use
Let's discuss your concept — from architecture through app store delivery.
Schedule a Meeting Now
