aether_skill

Overview

This skill is a full-stack orchestration system for building and operating AI VTubers (AITubers). It designs, implements, and monitors the real-time streaming pipeline from live chat ingestion through LLM response, TTS synthesis, avatar animation, and OBS output. The focus is low-latency, reliable presence with graceful degradation and observability.

How this skill works

Aether inspects project context and persona input, then generates a pipeline design with latency budgets, adapter interfaces, and stage-level configs. It implements chat integration (YouTube/Twitch), streaming LLM handling, chunked TTS adapters, lip-sync/expression mapping (Live2D/VRM), and OBS WebSocket automation. The system includes monitoring, alerts, and fallbacks so streams remain stable even under component failure.

When to use it

• Building a live AI-driven VTuber from persona to production stream
• Designing a low-latency TTS pipeline for interactive chat responses
• Integrating Live2D or VRM avatar control with lip sync and expressions
• Automating OBS scenes, encoding, and RTMP/SRT streaming configuration
• Auditing or tuning an existing AITuber pipeline for latency and reliability

Best practices

• Define an end-to-end latency budget (<3000ms target) and test against it
• Use adapter pattern for TTS and avatar engines to allow swaps without redesign
• Sanitize and moderate chat before LLM processing to protect viewers
• Implement streaming LLM + chunked TTS so audio can start before full response
• Add health monitoring and graceful degradation (TTS→text overlay, avatar→static)

Example use cases

• Design a pipeline for a new persona and hand off implementation specs to builders
• Build a VOICEVOX or SBV2 adapter and tune voice params per persona
• Integrate Twitch IRC and YouTube Live Chat with priority routing and moderators
• Automate OBS scene transitions on donations, polls, or persona cues
• Run a dry-run launch to validate latency, safety filters, and auto-recovery rules