Generating 3D Models for Game Characters from Hand-Drawn Illustrations

This paper proposes a 3D model generation method based on character illustrations specifically designed for game character production. Recent studies on single-view 3D model generation have demonstrated remarkable progress; however, due to the limited information provided by single-view images, the generated 3D models often suffer from a lack of details and incomplete geometric information. To solve these issues and improve the quality of 3D models, this paper presents a hand-drawn-illustration-based 3D model generation method that meets the high-quality and high-fidelity requirements of game character production. First, the proposed method enhances the 3D information of the input hand-drawn illustrations using Stable Diffusion. Subsequently, corresponding normal maps are generated through a pretrained diffusion model. Finally, the method uses a geometry-aware normal fusion algorithm to integrate the information from the input three-view images and the normal maps, enabling the generation of high-quality 3D models. Experimental results demonstrate that the proposed method outperforms existing single-view-based 3D model generation methods in terms of detail restoration and geometric consistency. This method effectively satisfies the demand for precise reproduction of design illustrations in game character production. Finally, this study also designed an evaluation algorithm for 3D model generation, providing a new approach for assessing the quality of AI-generated models.