Facial emotion detection has gained a significant importance in past years due to its unmatched potential applications in number of fields including to mental health monitoring, personalization in-real time, more friendly, and humanly computer-based interactions. In this paper, we aimed to perform a competitive analysis of performance of four different types of Convolution Neural Networks (CNN) architectures—Basic CNN, MobileNet, ResNet50, and finally VGG16. The models will be evaluated on their accuracy, recall, precision, F1-scores, and finally the number of epochs they were trained on, which essentially helps in finding the overall training time and compute efficiency. The work involves training all the models on a benchmarking dataset known as FER2013, known for its training efficient images and robust emotional expressions it offers. The models will be generally trained against the same parameters, to have a bias free comparison. Following the recognition of facial emotions, the second part is to recommend user a list of music based on the detected emotional state, aiming to enhance the user experience, and incorporating a GenAI-based chatbot that will be context aware and can help user is they are not feeling well and elevate the whole experience. The outcome of the paper is to compare and find human facial recognition task as well as personalizing the music recommendation system and proving mental health support, powered by GenAI. The performance analysis is that ResNet50 has outperformed all other models with accuracy of 76.71% in prediction, as compared to CNN with 62.72% accuracy, MobileNet with 74.81% accuracy, and VGG16 with 66.73% accuracy.