Harnessing the power of artificial intelligence and deep learning for creative expression, the proposed work presents a novel approach to music generation using Long Short-Term Memory (LSTM) networks. Aimed at synthesizing new compositions, the study employs a customized dataset of single-track piano music, encompassing both classical pieces and video game soundtracks converted into MIDI format. Through meticulous preprocessing, including the extraction of musical features using the Music21 library, the model’s architecture was carefully constructed with multiple LSTM layers and optimized with an Adamax optimizer for enhanced learning efficiency. With a significant training accuracy of 91.32% and a marked reduction in loss to 0.29, the model demonstrated proficiency in generating music that largely resonated with listeners, as evidenced by positive feedback from a quantitative survey. The Jaccard Similarity metric showed a score of 0, reflecting significant divergence and underscoring the model’s capacity to create novel outputs, potentially leading to new musical genres. A survey with 20 participants revealed that 15–16 reacted positively to the generated music, while 4–5 found the harmonium pieces lacking in soothing qualities, emphasizing the subjective nature of music appreciation. Despite some subjective feedback, the proposed work marks significant progress in generative AI for music composition, showcasing the unique contribution of this work which is its innovative use of LSTM networks to generate music that reflects existing styles while exploring new creative territories, thus pushing the boundaries of traditional composition and offering adaptable techniques for advancing AI-driven tools in the arts.