Generative Phomosaic with Structure-Aligned and Personalized Diffusion

2026-04-08

AI Summary

This work proposes the first generative framework for photomosaic synthesis, overcoming the limitations of traditional methods that rely on large precompiled tile libraries and color-based matching, which often fail to balance structural coherence with textural diversity. By leveraging a reference-guided diffusion model, the approach synthesizes tiles on-the-fly without requiring a pre-existing library, achieving both global structural alignment and rich local detail. The method integrates a low-frequency conditional diffusion mechanism with few-shot personalization techniques, transcending conventional tile-matching paradigms. It enables high-quality mosaic generation tailored to user-specified content or consistent artistic styles, significantly enhancing semantic expressiveness and visual personalization.

Abstract

We present the first generative approach to photomosaic creation. Traditional photomosaic methods rely on a large number of tile images and color-based matching, which limits both diversity and structural consistency. Our generative photomosaic framework synthesizes tile images using diffusion-based generation conditioned on reference images. A low-frequency conditioned diffusion mechanism aligns global structure while preserving prompt-driven details. This generative formulation enables photomosaic composition that is both semantically expressive and structurally coherent, effectively overcoming the fundamental limitations of matching-based approaches. By leveraging few-shot personalized diffusion, our model is able to produce user-specific or stylistically consistent tiles without requiring an extensive collection of images.

Problem

• photomosaic
• structural consistency
• diversity
• tile images
• color-based matching

Innovation

• generative photomosaic
• structure-aligned diffusion
• personalized diffusion
• low-frequency conditioning
• few-shot generation