<![CDATA[Generative Artificial Intelligence (Gen AI) technology is increasingly being used by creative professionals, including musicians and game developers. Many game developers now turn to open or paid music assets, but the variety of options is usually quite limited. This research aims to assist game developers in generating music assets in ABC notation format. The research methods include data collection in the form of ABC notation, data processing, model development, and metric evaluation. The data was collected by extracting ABC notation along with the characteristic musical components of each item. Data processing involved handling missing values and feature selection, while data preparation included labeling and tokenization. The model used was GPT-2 based on the Transformer architecture, pretrained on a general dataset. Integration of the model with ABC notation data was enhanced using Logits Processing to improve output control. The evaluation results show that Transformer technology can generate pitch patterns consistent with the validation data, with the EMD values concentrated in the range of 1.0–1.5 and an average of 1.60. Although there are some outliers and differences in pitch distribution between the validation data and generated results, the Horror genre with a Joyful mood and Excitement emotion achieved the highest combined fitness score of 0.528. The model still requires further refinement to produce more consistent pitch distributions. This research demonstrates the potential of Transformer technology in generating music assets for games, but further studies are needed to improve accuracy and consistency in the results.]]>
