One of the major areas of development in game engines over the past few decades has been graphical advancements allowing for the rendering of increasingly high quality environments in real time. While this has enabled new games to be increasingly visually impressive, there is increasing interest in modifying and redeveloping older video games to run on these advanced rendering frameworks.

Due to graphical limitations of older games and game engines, the art assets they utilised were often low resolution and simplistic to fit the capabilities of consumer hardware of the time. Due to these limitations, simply using old game art assets in modern game engines will not result in a game that looks as good as games designed for a modern engine from the start, requiring the assets to be recreated either manually or through automatic processes.

Recent advancements in Artificial Intelligence (AI) and Deep Learning, including but not limited to Generative AI, offer new user directed solutions for improving existing assets or creating new assets from scratch, or using various text and image based sources as input. This research aims to explore how AI can be used to upgrade older game textures which is ideal for modern rendering techniques. Specifically, this project will be investigating using AI to take diffuse textures with baked lighting which is common in games, and using these as a basis to create textures required for Physically-Based Rendering (PBR). We will be focusing on AI models which can create normal, roughness, specular and diffuse textures for video games.