Meta unveils Muse Realtime Avatar: synced voice-and-video AI that responds in under a second
Scobleizer · x · 2026-09-24
Meta announced Muse Realtime Avatar, paired with Muse Realtime Voice: any character can now hold a live conversation, animated in sync with its speech, responding in under a second for unlimited-duration chats.
Engineering lead kakemeister explained the team rebuilt Meta's realtime AI inference stack since joining from Waveforms two years ago, calling realtime multimodal inference one of the hardest AI infra problems—spanning model/hardware optimization, LLM orchestration, and the RTC stack—with compute-bound diffusion video models posing entirely new latency and cost challenges. Chief AI Officer Alexandr Wang billed it as state-of-the-art.
Related event: Meta Launches Muse Realtime Avatar for Sub-Second AI Avatars(5 posts)→
More from Infra
- AI Infra Startups Modal and Baseten in Funding Talks, Bloomberg Reports — dinabass · 2026-09-24
- CUbiC Paper Outlines Edge-to-Cloud Connectivity for AI Infrastructure — jwt0625 · 2026-09-24
- Hunyuan Research: Batch-Size Scaling with LR Retuning Boosts PPO Throughput 2.29x — TencentHunyuan · 2026-09-24
- Alchemy adds opt-in Cloudflare Access protection for its state store, with CI service tokens — samgoodwin89 · 2026-09-24
- A Ready-to-Use Prompt That Makes Your Agent Audit Its Own API Bills — gethackteam · 2026-09-24
- ~50us per kernel launch possible, but only by forking a custom single-model inference stack — AlpinDale · 2026-09-24