OmniVBench and the associated Omni-R2V dataset provide a large-scale evaluation framework for reference-to-video models. The benchmark specifically targets the ability of models to preserve, disentangle, and correctly route diverse types of reference information.
HOW THIS AFFECTS YOU
●
researcherYou can now evaluate how well your video models handle complex, multi-factor reference control.
●
designerThis allows for more granular testing of how creative references are translated into video.