Overview
- The Copilot Labs experiment offers roughly 40 stylized human avatars for real-time voice conversations in a limited rollout across the US, UK and Canada.
- Microsoft says the feature uses VASA-1 to generate facial expressions, head movements and lip-sync from a single image without complex 3D modeling.
- Access is restricted to users 18 and older with clear AI disclosures, existing Copilot safety filters and session caps observed at about 20 minutes.
- Early hands-on reports describe unsettling behavior such as a fixed stare, auto-greetings by name, inconsistent accents across sessions and brief audio dropouts.
- Portraits follows a prior Copilot avatars preview and arrives as rivals push similar features, with Microsoft calling this a cautious, feedback-driven test.