The first benchmark for multi-factor sequential disentanglement representations, introduces a novel method, and leverages Vision-Language Models to automate annotation and evaluation—enabling scalable, label-free workflows.