More Information

Background

Many games, such as survival titles, rely on procedurally generated maps to create replay value without presenting identical levels every time. Methods have long existed to automatically generate maps while adhering to certain rules (e.g., ensuring the map remains playable or fits a specific style). Newer generative AI models, such as LLMs or diffusion models, promise more flexible and expressive generation. It remains to be seen how well they are actually suited for playable maps that must adhere to specific constraints and how they compare to traditional methods.

Research Question

This paper is based on a structured literature review. The first step involves reviewing traditional methods for automatic, rule-based card generation; the focus then shifts to generative AI. The central question is which methods are currently gaining traction in this field and where generative AI offers actual added value compared to traditional approaches. A prototype implementation is also planned.