Abstract

Image tiling -- the seamless connection of disparate images to create acoherent visual field -- is crucial for applications such as texture creation,video game asset development, and digital art. Traditionally, tiles have beenconstructed manually, a method that poses significant limitations inscalability and flexibility. Recent research has attempted to automate thisprocess using generative models. However, current approaches primarily focus ontiling textures and manipulating models for single-image generation, withoutinherently supporting the creation of multiple interconnected tiles acrossdiverse domains. This paper presents Tiled Diffusion, a novel approach thatextends the capabilities of diffusion models to accommodate the generation ofcohesive tiling patterns across various domains of image synthesis that requiretiling. Our method supports a wide range of tiling scenarios, from self-tilingto complex many-to-many connections, enabling seamless integration of multipleimages. Tiled Diffusion automates the tiling process, eliminating the need formanual intervention and enhancing creative possibilities in variousapplications, such as seamlessly tiling of existing images, tiled texturecreation, and 360{\deg} synthesis.