Emu Video by Meta

Emu Video employs novel factorization separating content, motion, and style for unprecedented video realism. Researchers advance generative video while creators produce professional footage without cameras. Joint audio-visual modeling synchronizes sound naturally. Decoupled generation maintains subject identity across motion extremes while style transfer applies cinematic looks consistently. Long-sequence coherence spans minutes of footage with consistent physics and lighting. Multi-modal conditioning supports image+text inputs. Research demonstrates scaling laws for video realism beyond images. Human evaluations confirm superiority over prior diffusion models. Technical report details factorization architecture enabling future improvements. Free research release includes pretrained models and inference code. Enterprise compute required for production. Primarily advances fundamental video generation research. Explore AI tools.