Existing human--object interaction (HOI) video generation methods are largely limited to offline short-video generation with complex driving conditions, making them unsuitable for real-time interactive applications. We present \emph{StreamHOI}, a low-latency streaming framework f…
Generating lifelike facial animation for dyadic conversations requires reconciling high-level cognitive intent with precise low-level motor reflexes, yet existing methods fall short in the semantic understanding of dialogue context and in precise dynamic control. In this paper, w…