Static thumbnails on popular video platforms such as YouTube, Rumble, Crackle, and Netflix often fail to convey the essence of video content, resulting in a mismatch between viewer expectations and the actual material. This paper presents an innovative solution that utilizes Artificial Intelligence (AI) to transform conventional static thumbnails into dynamic highlight clips, providing viewers with an engaging and concise overview of the video. Our proposed model employs convolutional neural networks for effective feature extraction, which is subsequently processed through a Bidirectional Long Short-Term Memory (BiLSTM) network enhanced with an attention mechanism. This combination enables the identification of critical moments within videos, allowing the generation of short video snippets that serve as intelligent thumbnails. To demonstrate the effectiveness of our approach, we conduct experiments using benchmark datasets, including TVSum and SumMe, which feature a diverse range of video content. The model's performance is evaluated using key engagement metrics, such as accuracy, F-Score, and Click-Through Rate (CTR), to assess the impact of dynamic previews compared to traditional static thumbnails. Our results reveal significant enhancements in user interaction, with an F-Score of 0.579 for TVSum and a notable increase in CTR for the dynamic thumbnails. These findings underscore the advantages of AI-generated video highlights in enhancing content discoverability and viewer engagement on video-sharing platforms.