Work

Continuous Level Generation Using Conditional Diffusion Models

Year: 2024
Type: article
Abstract: Procedural content generation came into sight since the 1970s.The increasing complexity and scale of modern games, coupled with the desire for diverse and replayable experiences, have propelled the game development community toward embracing procedural techniques, therefore procedural content generation has gained prominence in the field of game development and simulations for its ability to generate vast and varied content automatically, like texture, foliage, dungeon, terrain, landscape, etc., improving the productivity of 2D/3D games by a large amount.One of the most intriguing of procedural generation is level generation.This departure from handcrafted level design not only addresses the demands for positioning visual elements for a number of levels but also introduces unpredictability and meanwhile captures game mechanics, enhancing player immersion and engagement.Diffusion model has been one of the most powerful generative models in recent years, it has not only found relevance in academic research but has also made significant contributions to practical domains such as image inpainting, super-resolution, image generation, and more.However, research on its application on level generation is still an uncharted area and is in need of comprehensive investigation and experiments.This research explores the application of diffusion models with a novel continuous, condition-based approach and conducts experiments to generate Super Mario Bros (SMB) level and proves its capability to generate levels with high quality by getting evaluated quantitatively and qualitatively.
Author: XIAO, Linyun
Language: English
Open Access status: green
OpenAlex ID: W6990187113
FWCI:
Cites: 30
Cited by:
Related to: 10