PhotoQuilt: Training-Free Arbitrary-Resolution Photomosaics via Bootstrapped Tiled Denoising

2026-06-29

AI Summary

High-resolution photomontage generation is often hindered by substantial computational costs and the inherent trade-off between global structural coherence and local textural fidelity. This work proposes a training-free, guidance-driven tiled denoising framework that first establishes a globally consistent layout at low resolution and then refines local details through latent-space upsampling combined with noise re-injection, enabling independent denoising within fixed tiles to preserve photorealism. By innovatively integrating global guidance with a tiled generation mechanism, the method circumvents both model training and the computational overhead of cross-attention operations, allowing efficient synthesis of high-resolution collages that maintain both structural consistency and fine-grained realism—significantly outperforming existing baselines across arbitrary output scales.

Abstract

Photomosaics are large images whose local regions are seen as independent tiles while their overall arrangement forms a coherent scene. Generating them at high resolution, with every tile convincing in its own right, is computationally expensive, since the canvas must hold many detailed tiles at once. We present PhotoQuilt, a training-free framework that generates photomosaics at arbitrary resolution. Diffusion models struggle to satisfy both scales at once, as direct high-resolution generation is costly and tends toward one smooth image rather than a mosaic, while patch-based tiling keeps local detail but loses global structure. PhotoQuilt resolves this with a bootstrapped tiled denoising procedure. We first produce a global composition at low resolution to fix the layout, then upscale it in latent space and re-inject noise to restore generative capacity. Denoising proceeds within fixed tiles, so each forms its own image while the shared global structure holds them in one layout. Because tile generation is handled separately, PhotoQuilt scales to large canvases without quadratic attention cost. Experiments show that PhotoQuilt outperforms current baselines on both global structure and local realism.

Problem

• photomosaics
• arbitrary-resolution
• diffusion models
• global structure
• local realism

Innovation

• photomosaic
• training-free
• tiled denoising
• diffusion models
• arbitrary resolution