Leveraging Generative AI to Create Themed Assets for Games : A Study on Narrative and 3D Asset Creation With AI

Developing games includes various steps, from creating a game concept to art creation and implementation. This process requires skilled professionals and a lot of time and resources. This thesis explores how generative artificial intelligence (AI) can assist in creating themed game assets, specifically short narrative stories and 3D models, for use in an augmented reality (AR) treasure-hunting game primarily meant for kids up to the age of ten. A web-based interface was developed to allow developers to generate a complete set of assets based on a chosen theme. Alongside the web application, various generative AI models' performance was evaluated, including GPT-4o and Gemini 2.0 for creating stories, and Hunyuan3D 2.0 and Luma AI Genie for creating 3D assets. The generated assets were evaluated through both quantitative and qualitative methods. Readability metrics, including Flesch-Kincaid and New Dale-Chall scores, were used to assess the stories' suitability for the target audience. For the 3D models, manual inspection and a user survey were conducted to evaluate the perceived quality and usability in a game setting. Due to the novelty of the 3D modeling tools and the lack of standardized methods for assessing the generated models, this study used experimental evaluation metrics based on methods for similar technology, such as text-to-image generation. The findings showed that while current generative AI models can produce themed assets that are visually compelling and reasonably easily read stories, there are notable limitations. The results suggest that AI can assist in asset creation, but today's models need more refinement to produce plugin-ready assets.