Roadmap: photorealistic / video avatar visual layer

[ykshv]

Goal

Design and implement a replaceable visual layer for a more realistic avatar path.

Scope

• Keep the current VRM path working.
• Define the boundary for TTS audio -> audio/video renderer -> WebRTC video track.
• Preserve session_id, turn_id, generation_id, branch_state, seq, and pts_ms semantics.
• Document latency, consent, licensing, and GPU implications.

Acceptance

• A design note or implementation plan exists.
• Stale video/avatar frames cannot survive a generation change.
• The README can clearly distinguish VRM mode from future photorealistic/video mode.