LPM 1.0은 실시간으로 말하고 노래하며 듣고 반응하고 표정짓는 비디오 기반 캐릭터 퍼포먼스 모델이다. 풀-듀플렉스 대화, 정체성 일관성의 무한 길이 생성, 감정 표현과 정확한 립싱크를 갖춰 대화형 에이전트·라이브 스트리밍 캐릭터·게임 NPC의 시각 엔진으로 활용된다. 데이터 파이프라인·베이스·온라인 모델·스트리밍 추론 최적화를 통해 높은 성능과 자연스러움을 목표로 한다.

https://x.com/AilingZeng81332/status/2042464987529298244

#videogeneration #realtime #virtualavatars #lipsync #multimodal

Exclusive TrailblaXR interview: Virbela

https://instinctivelyreal.com/2023/11/22/virbela-interview/

#AI #DeepFakes #Doppelgangers #VirtualAvatars #BigData: "In the ongoing, historic Hollywood writers’ and actors’ strike, a major point of contention is how AI and emerging technologies threaten these workers’ creative and intellectual rights. In particular, actors are raising concerns about “digital doubles” — the AI-powered ability to reproduce an actor’s likeness, voice, and appearance, which studios could presumably use (and profit from) in perpetuity, without the actor’s direct participation. Concerned actors are grappling with a question that strikes at the heart of their profession: What are the consequences of having a digital double when your persona is the foundation of your craft and career?"

https://points.datasociety.net/doppelgangers-in-a-digital-world-48aa512b69bd