Video captioning, the task of generating natural language descriptions for video content, plays a pivotal role in bridging the gap between visual understanding and language generation. This study introduces a novel approach that leverages Energy-Based Models (EBMs) combined with Energy Distance (ED) to optimize the alignment between video features and captions. The proposed VC-EBM-ED framework emphasizes compatibility by minimizing energy discrepancies between correct and incorrect video-caption pairs, allowing the model to learn nuanced relationships effectively. Experiments conducted on the MSR-VTT dataset demonstrate the model’s ability to generate captions that capture both the overall context and specific actions within videos. Results highlight the framework’s potential for generalization across unseen data, underscoring its robustness and adaptability. While the model achieves promising performance, further exploration into handling longer and more complex video scenarios presents an exciting avenue for future research. The VC-EBM-ED framework offers a foundation for advancing multimodal learning and unlocking new possibilities in video understanding and captioning applications.