Procedural Content Generation via Generative Artificial Intelligence

TL;DR

This survey analyzes how generative AI, notably GANs, diffusion models, and transformers, is transforming procedural content generation in video games. It covers content types from 2D levels and 3D terrains to art, animation, narrative agents, and audio, emphasizing data scarcity as a fundamental bottleneck and exploring strategies to cope with limited training data. Through a survey of landmark works, it highlights the shift from rule-based PCG to data-driven, multi-stage, and multi-modal approaches, and discusses practical concerns such as playability, realism, compute costs, and validation. The paper concludes with future directions toward higher-quality, more controllable content, NL-driven prompts, and user-generated content ecosystems that could reshape game development workflows.

Abstract

The attempt to utilize machine learning in PCG has been made in the past. In this survey paper, we investigate how generative artificial intelligence (AI), which saw a significant increase in interest in the mid-2010s, is being used for PCG. We review applications of generative AI for the creation of various types of content, including terrains, items, and even storylines. While generative AI is effective for PCG, one significant issues it faces is that building high-performance generative AI requires vast amounts of training data. Because content generally highly customized, domain-specific training data is scarce, and straightforward approaches to generative AI models may not work well. For PCG research to advance further, issues related to limited training data must be overcome. Thus, we also give special consideration to research that addresses the challenges posed by limited training data.