MoCha: Towards Movie-Grade Talking Character Generation

"Two distinct streams of tears trail down her cheeks as she speaks with an angry expression…" "A medium shot of a man interacting warmly with an elephant. the man talks to the camera…" "A tilt up shot of a man standing in a dimly lit room, speaking to the camera…" Action Control Multi-Character Turn-based Talk Talking Character "Close-up shot of a doctor in a white lab coat over blue scrubs, speaking…" Emotion Control Figure 1: MoCha is an end-to-end dialogue-centric video generation model that takes only speech and text as input, without requiring any auxiliary conditions.