The Role of Generative AI in Agricultural Game Assets Production: A Survey

Abstract

The growing demand for high-quality visual assets in the gaming industry has created challenges in producing diverse and realistic agricultural-themed content. Traditional techniques, such as procedural content generation (PCG) and early deep learning models—including Autoencoders, Variational Autoencoders, and Deep Convolutional GANs—often struggle with limitations in resolution, realism, and diversity. This study aims to explore how Generative Adversarial Networks (GANs), particularly StyleGAN2, can address these challenges in generating 2D assets for farming games. A systematic literature review was conducted using databases such as Scopus and Google Scholar, covering publications from the last five years. The selection criteria included studies focusing on generative models for visual game asset creation, with an emphasis on domains related to agriculture or the environment. The review highlights StyleGAN2’s style-based architecture, which enables fine-grained control over sprites, textures, and environmental elements, leading to more realistic and customizable assets. Key contributions of this work include identifying current technical strengths, outlining socio-economic implications, and discussing practical challenges such as dataset availability and evaluation standards. The findings suggest opportunities for hybrid procedural–AI approaches, domain-specific datasets, and the expansion of content toward dynamic and interactive elements. By consolidating these insights, the paper offers guidance for both researchers and practitioners on leveraging generative AI for the realistic and diverse production of agricultural game assets.