OmniVBench introduces a benchmark and large‑scale dataset for omni reference‑to‑video (R2V) generation, addressing the gap in existing benchmarks that cover limited reference types and evaluate only holistic consistency. It provides diverse reference compositions and evaluation protocols that assess whether individual reference factors are preserved, disentangled, and properly routed. The work aims to enable rigorous measurement of versatile reference control in video synthesis.