Year: 2025
Type: dissertation
Abstract: The use of Artificial Intelligence (AI) for assisted item generation in digital games has emerged as a promising approach to optimizing the visual content creation process. This study investigates the application of generative models based on diffusion neural networks, emphasizing the integration of Stable Diffusion XL and auxiliary techniques such as ControlNet, T2I Adapter, and PidiNet for contour detection. The primary objective is to develop and evaluate an automated system capable of generating artwork for game items based on reference images. The adopted methodology includes the implementation of an image generation pipeline in Python, utilizing specialized libraries for neural processing and execution on accelerated hardware. The process involves contour extraction, the application of diffusion models for visual refinement, and the quantitative and qualitative analysis of the obtained results. This study contributes by proposing a structured workflow for the automated creation of visual elements in games, reducing production time and increasing the aesthetic variability of the generated items. The results indicate that the AI-based approach surpasses traditional techniques in terms of flexibility and efficiency, enabling the creation of high-quality images with minimal human intervention. It is concluded that the combination of generative models and image segmentation techniques represents a significant advancement in game development, opening new possibilities for the automation of graphical production.
Author: Olenik, Mateus Marochi
Institution: Universidade Tecnológica Federal do Paraná
Language: Portuguese
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7155463684
