OmniXRef Enables Multimodal Multi-Reference Video Editing
October 14, 2026
OmniXRef is a unified framework using a single-stream diffusion transformer to edit video using combined text, image, and video references. It utilizes a reference-bank interface and R^2-RoPE to generalize across variable numbers of multimodal inputs.
HOW THIS AFFECTS YOU
●
builderThis framework provides a unified architecture for complex multimodal video editing tasks.
●
designerYou can use multiple reference types to exert finer control over video generation.