Match-and-Fuse: Consistent Generation from Unstructured Image Sets

2025-11-27

AI Summary

This work addresses zero-shot, consistency-controllable image generation from unstructured image sets—collections sharing common visual elements but exhibiting diverse viewpoints, timestamps, and backgrounds. We propose the first training-free set-to-set generation framework that models each image set as a graph, establishes cross-graph correspondences via dense feature matching between image pairs, and achieves local consistency and global coherence through cross-image feature fusion and a shared canvas prior. The method requires no mask annotations, human supervision, or video-sequence constraints. Guided by text prompts, it jointly synthesizes novel image sets with diverse viewpoints and semantically coherent content. Quantitative and qualitative evaluations demonstrate state-of-the-art performance in both structural consistency and visual fidelity, significantly advancing controllable creative synthesis from unordered image collections.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Scalable techniques for the creation, curation, publication, maintenance, and consumption of large, Web-based, structured, reusable, knowledge graphs and ontologies
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

We present Match-and-Fuse - a zero-shot, training-free method for consistent controlled generation of unstructured image sets - collections that share a common visual element, yet differ in viewpoint, time of capture, and surrounding content. Unlike existing methods that operate on individual images or densely sampled videos, our framework performs set-to-set generation: given a source set and user prompts, it produces a new set that preserves cross-image consistency of shared content. Our key idea is to model the task as a graph, where each node corresponds to an image and each edge triggers a joint generation of image pairs. This formulation consolidates all pairwise generations into a unified framework, enforcing their local consistency while ensuring global coherence across the entire set. This is achieved by fusing internal features across image pairs, guided by dense input correspondences, without requiring masks or manual supervision. It also allows us to leverage an emergent prior in text-to-image models that encourages coherent generation when multiple views share a single canvas. Match-and-Fuse achieves state-of-the-art consistency and visual quality, and unlocks new capabilities for content creation from image collections.

Problem

• Generates consistent image sets from unstructured collections
• Preserves shared content across varying viewpoints and times
• Ensures global coherence without training or manual supervision

Innovation

• Graph-based set-to-set generation for image consistency
• Fusion of internal features using dense correspondences without masks
• Leveraging emergent prior in text-to-image models for coherence