<p>Automatic music generation (MG) is essential for video games, films, and other interactive media, especially when continuous soundtracks are needed to captivate audiences. However, developing music that is realistic, expressive, and stably stylistic is still challenging. Traditional rule-based or statistical methodologies often produce repetitive or mechanical music characterized by none of human feelings found in musical compositions. To design and develop an automatic MG model to learn musical patterns and produce music of high quality. This paper proposes an automatic MG based on Long Short-Term Memory (LSTM) networks, called MGLSTM. The model implements a distinct set of Musical Instrument Digital Interface (MIDI) files in classical, pop, and jazz genres. The music21 library processes the MIDI files to extract the subsequent chain of the notes and chords. Each musical entity, note, and chord, is converted into a numerical form appropriate for deep learning. The MGLSTM implements a sequence-to-sequence approach to train the model to learn relationships and patterns in the musical data. The MGLSTM also optimizes the Adam model for training to increase convergence speed and model performance. Once trained, the model is capable of the automatic generation of new musical compositions that define the style of the training dataset. Experimentally the MGLSTM produced a note prediction accuracy of 85%. In listening tests, the majority of participants all attributed the generated music as natural and musically pleasing. These findings underscore the efficacy of the LSTM models in generating high-quality music that adheres to styles and genres. The method proposed in this work provides a robust automatic music content generation approach with a higher structural cohesion and enjoyment value than traditional methods.</p>