Digital Game Development Using Large Language Models (LLMs): An Exploratory Study

Introduction: Large Language Models (LLMs) are powerful tools for automating tasks like documentation, code generation, and prototyping in computer science, but their integration into game development pipelines is an opportunity by, also a challenge. Objective: This paper presents the development and implementation of PromptingGameCraft (PGC), a tool that uses Large Language Models (LLMs) to automate key steps in digital game development. The tool takes a Game Design Document (GDD) as input and automatically generates a Game Design File (GDF) in JSON format, along with a custom class diagram, directory and file structure, and game code. Methodology: The architecture was implemented through a web interface connected to the DeepSeek-reasoner model API hosted on Google Cloud. As a proof of concept, a 2D ball-catching game with progressive difficulty was developed. Results: The automated generation process demonstrated efficiency in the transition from design to code, promoting modular organization, logical clarity, and reusability. In addition to productivity and standardization, PGC has the potential to democratize access to game development in educational, training, and community contexts. By enabling the accessible transformation of ideas into working prototypes, it promotes creative expression, supports active learning, and enhances participation among diverse groups.