Towards Generalized Multi-Image Editing for Unified Multimodal Models

2026-01-09

arXiv.org

AI Summary

This work addresses the challenge that existing unified multimodal models struggle to maintain visual consistency and accurately disentangle visual cues from multiple input images in multi-image editing tasks. To overcome this limitation, the authors propose a scalable multi-image editing framework that introduces learnable latent separators to decouple conditional information from individual reference images and incorporates sinusoidal index encoding to enable identity-aware generalization across an arbitrary number of inputs. The effectiveness of the approach is validated on a high-quality, reverse-engineered multi-image editing benchmark dataset. Experimental results demonstrate that the proposed framework significantly outperforms current methods in semantic consistency, visual fidelity, and cross-image fusion, highlighting its superior capability in achieving both consistent editing and robust generalization.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Unified Multimodal Models (UMMs) integrate multimodal understanding and generation, yet they are limited to maintaining visual consistency and disambiguating visual cues when referencing details across multiple input images. In this work, we propose a scalable multi-image editing framework for UMMs that explicitly distinguishes image identities and generalizes to variable input counts. Algorithmically, we introduce two innovations: 1) The learnable latent separators explicitly differentiate each reference image in the latent space, enabling accurate and disentangled conditioning. 2) The sinusoidal index encoding assigns visual tokens from the same image a continuous sinusoidal index embedding, which provides explicit image identity while allowing generalization and extrapolation on a variable number of inputs. To facilitate training and evaluation, we establish a high-fidelity benchmark using an inverse dataset construction methodology to guarantee artifact-free, achievable outputs. Experiments show clear improvements in semantic consistency, visual fidelity, and cross-image integration over prior baselines on diverse multi-image editing tasks, validating our advantages on consistency and generalization ability.

Problem

• multi-image editing
• visual consistency
• image identity disambiguation
• unified multimodal models
• cross-image reference

Innovation

• learnable latent separators
• sinusoidal index encoding
• multi-image editing
• unified multimodal models
• visual consistency