To address the limitations of existing evaluation protocols in reference-to-video (R2V) generation, researchers have introduced OmniVBench and the Omni-R2V Dataset. Current benchmarks often fail to assess whether models can properly disentangle and route specific reference factors, prompting the development of OmniVBench. This new benchmark covers 7 task families and 18 fine-grained tasks, including content, motion, style, structure, narrative, and multi-reference settings. It utilizes a factor-grounded evaluation method consisting of 12,172 case-specific checklist items to determine if intended reference factors are faithfully preserved and correctly bound to their targets. To support training for these complex tasks, the authors released the Omni-R2V Dataset, which provides 340,000 processed training samples derived from a large-scale corpus of professional video footage. The dataset utilizes specialized task-specific pipelines for constructing reference-target pairs, offering a scalable methodology for data production. Through extensive testing of various open- and closed-source R2V models, the study identifies significant performance gaps across different task families and evaluation dimensions, highlighting the current technical deficiencies in achieving precise, multi-faceted reference control in video generation.