Survey maps how generative AI is reshaping procedural content creation

Researchers have completed a comprehensive examination of how generative AI techniques are being applied to procedural content generation (PCG)—the automated creation of digital assets and experiences. The survey, submitted to arXiv in July 2024, tracks the evolution of machine learning approaches in this domain and explores emerging applications across diverse content types.

Applications across multiple content domains#

Generative AI systems are now being adapted to produce a wide range of assets and interactive elements. The research documents successful applications in creating game terrains, in-game items, and even narrative storylines. These applications represent a shift from earlier, rule-based procedural generation methods, offering designers new tools to scale content creation while maintaining creative control over the final output.

The data challenge ahead#

Despite its promise, generative AI for PCG faces a significant bottleneck: the need for massive volumes of training data. Because game worlds and interactive content are highly specialized and domain-specific, relevant training datasets remain limited. The survey emphasizes that off-the-shelf generative AI approaches often fall short when applied directly to PCG tasks, requiring researchers to develop tailored solutions that work within these data constraints.

Research priorities#

The survey identifies the scarcity of domain-specific training data as a critical challenge that must be addressed for the field to mature. Researchers are increasingly focusing on techniques that can work effectively with limited datasets—an area the survey highlights as essential for advancing procedural content generation research going forward.