In 2025, Lima, Peru, enacted legislation mandating that all motorcyclists wear certified helmets and reflective vests displaying their license plate number in response to increased armed robberies by criminal gangs. This study develops a YOLOv8 object-detection system using real-world data and synthetic video data generated with Google’s Veo 2 and Veo 3 models to identify compliance with this new law. Manually recording missing vests or reading license plates in real-world traffic footage is both costly and time-consuming. We hypothesize that a computer vision object detection model can operate in real-time at traffic lights to automatically notify authorities of non-compliant riders. We synthesized short clips of virtual riders under various lighting conditions, camera angles, weather conditions, and urban backdrops depicting both compliant and non-compliant scenarios. We strengthened our detection system using image-to-video prompting to generate more diverse and realistic synthetic training data. These approaches enabled us to train and evaluate three YOLOv8 models using different data strategies to assess how synthetic data quality and diversity affect detection accuracy for real-world deployment.