This research presents an advanced approach to content moderation in Instagram Reels, aiming to protect young adults from exposure to inappropriate material. Central to this research are three models: a Convolutional Neural Network (CNN) for extracting spatial features from videos, and a Long Short-Term Memory (LSTM) model for analyzing temporal data, together achieving an accuracy of 89.99% in detecting unsuitable content. Additionally, the study validates the effectiveness of a specialized CNN-based model for smoking detection, which attains a test accuracy of 90.59% and a Ensemble model of both incorporated together achieving an accuracy of 74%. These models showcase the importance of combining spatial and temporal analyses for effective video content moderation. The research outcomes not only enhance understanding of digital content moderation strategies but also contribute to improving user safety and experience on social media platforms like Instagram Reels.