This document proposes a content-based video recommendation system using low-level visual features to address the new item problem. It extracts features like average shot length, color variance, and motion from video content to classify videos into genres. An evaluation of the system used 120 videos across 4 genres and achieved a classification accuracy of 73.33%. The results showed this approach performed better than baselines in addressing the new item problem. Future work could analyze larger datasets and incorporate additional modalities like audio.