Generic Event Boundary Detection (GEBD) is a relatively new but essential task in video understanding, aimed at detecting generic and class-agnostic boundaries between semantically different parts of videos. This study introduces a novel zero-shot and unsupervised method for GEBD using a graphical representation of videos and the normalized cut algorithm. Our method leverages off-the-shelf lightweight models, specifically ResNet50 and DINOv2, and requires no additional training or fine-tuning, making it highly efficient in time and compute. Inspired by similar tasks in 2-D computer vision, we employ the normalized cut algorithm as a novel and simple approach to detect event boundaries in time-series data such as videos. Our approach achieves state-of-the-art F1@0.05 scores of 0.732 on the Kinetics-GEBD dataset and 0.398 on the TAPOS dataset, surpassing existing accuracy and computational efficiency methods. This study underscores the potential of using normalized cut for GEBD and highlights its scalability for future applications in video understanding.