OmniWeaving: Towards Unified Video Generation with Free-form Composition and Reasoning
OmniWeaving proposes a unified video generation model that handles free-form video creation by combining multimodal understanding and reasoning. It integrates a Multimodal Large Language Model, a Multimodal Diffusion Transformer, and a VAE to temporally bind text, images, and videos. This allows the model to infer complex user intentions for sophisticated video creation.