STARFlow-V is the first normalizing flow-based causal video generator demonstrating that normalizing flows can match video diffusion models in visual quality while offering end-to-end training, exact likelihood estimation, and native multi-task support across T2V/I2V/V2V generation.