Modern digital technology emphasizes customization and automation more than ever before. Conventional music recommendation systems depend on user preferences, historical data, and collaborative filtering, often neglecting current emotional states. To make up for this shortcoming in literature, a real-time music recommendation system fueled by transfer learning for facial emotion detection has been designed with EfficientNet and MobileNet models. In our method, we focused particularly on these two models due to the balance they strike on accuracy, efficiency, and suitability for real-time applications. The system captures live facial expressions through a webcam and applies fine-tuned pre-trained models on the acquired input to determine the emotional state of a user. The playlists shall be generated based on the detected emotion, where user feedback on the previous recommendations builds a more relevant list for future recommendations. This approach improves user experience by giving suggestions for music that match everyone else’s state of emotion, successfully combining deep learning with affective computing. Compared to existing systems, it improves accuracy, broadens the number of detectable emotions, and offers tunes streaming recommendations more personalized than before.