MICON-Bench: Benchmarking and Enhancing Multi-Image Context Image Generation in Unified Multimodal Models

2026-02-22

AI Summary

Existing benchmarks struggle to evaluate unified multimodal models on multi-image contextual generation tasks, particularly in terms of cross-image composition, contextual reasoning, and identity preservation. To address this gap, this work proposes MICON-Bench—the first systematic evaluation benchmark tailored for multi-image contextual generation—encompassing six core tasks. It introduces two key technical contributions: an automatic semantic and visual consistency verification framework based on multimodal large language models (MLLMs), and a training-free, plug-and-play Dynamic Attention Rebalancing (DAR) mechanism. Experimental results demonstrate that MICON-Bench effectively uncovers limitations in current models, while DAR significantly enhances the consistency and contextual coherence of generated outputs and reduces hallucinations.

Technology Category

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

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Recent advancements in Unified Multimodal Models (UMMs) have enabled remarkable image understanding and generation capabilities. However, while models like Gemini-2.5-Flash-Image show emerging abilities to reason over multiple related images, existing benchmarks rarely address the challenges of multi-image context generation, focusing mainly on text-to-image or single-image editing tasks. In this work, we introduce \textbf{MICON-Bench}, a comprehensive benchmark covering six tasks that evaluate cross-image composition, contextual reasoning, and identity preservation. We further propose an MLLM-driven Evaluation-by-Checkpoint framework for automatic verification of semantic and visual consistency, where multimodal large language model (MLLM) serves as a verifier. Additionally, we present \textbf{Dynamic Attention Rebalancing (DAR)}, a training-free, plug-and-play mechanism that dynamically adjusts attention during inference to enhance coherence and reduce hallucinations. Extensive experiments on various state-of-the-art open-source models demonstrate both the rigor of MICON-Bench in exposing multi-image reasoning challenges and the efficacy of DAR in improving generation quality and cross-image coherence. Github: https://github.com/Angusliuuu/MICON-Bench.

Problem

• multi-image context generation
• unified multimodal models
• cross-image composition
• contextual reasoning
• identity preservation

Innovation

• MICON-Bench
• Dynamic Attention Rebalancing
• Multi-Image Context Generation
• MLLM-driven Evaluation
• Unified Multimodal Models