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

TL;DR

MICON-Bench introduces a new comprehensive benchmark for multi-image context generation in multimodal models, along with a novel evaluation framework and a dynamic attention mechanism to improve coherence and reduce hallucinations.

Contribution

This work presents MICON-Bench, a new benchmark for multi-image reasoning, an MLLM-driven evaluation framework, and a plug-and-play attention rebalancing method for enhanced image generation.

Findings

MICON-Bench effectively exposes multi-image reasoning challenges.

DAR improves coherence and reduces hallucinations in image generation.

The evaluation framework automates semantic and visual consistency verification.

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.