Ex-Google AR founder launches Atmee v2: real-time avatar model claims 50x speed, 100x cheaper
eptwts · x · 2026-11-05
yuchenping, whose AR startup was acquired by Google, launched Atmee v2, a real-time avatar video foundation model aiming to make live video agents feel truly life-like.
Key points:
- Any character from 1 image + 10s of audio
- Contextual expressions, gestures, and environment interaction
- Live video calls, voice-only, or video generation modes
- A video foundation model built to excel at characters
- Claims up to 50x faster and 10-100x cheaper than general video models like Seedance 2.5 and Wan 3.0
He argues text/voice are the DOS era of AI interfaces; face-to-face video is next. Use cases: tutors, sales coaches, support agents, companions, game characters, living worlds.
Related event: Atmee v2 Enables Realistic Real-Time Video Calls from One Photo(2 posts)→
More from Multimodal
- 2-person lab trains 2B text-to-image model from scratch for $20K with Pyramid-JiT — generativist · 2026-11-05
- LFM3-Omni-2B ported to run fully offline on iPhone at ~80 tok/s via custom WebGPU kernels — victormustar · 2026-11-05
- Google ships Voice Remixing for Gemini 3.8 Flash TTS with 2,000+ preset voices and prompt-based control — GoogleDeepMind · 2026-11-05
- Fixing the 'Plastic Wax' problem: a technical workflow for believable AI influencer skin — NameAdditional3979 · 2026-11-05
- Glif signs AI influencer Claudia Popstar (44M views) to its AI modeling agency — HeyAmit_ · 2026-11-05
- AI artist AMARA ctk100 racks up millions of Spotify streams as listeners vibe daily — kyliebytes · 2026-11-05