UFO: Chain-of-Evaluation for Omni-Condition Alignment in Multi-Modal Image Generation

Multi-modal image generation, particularly subject-driven customization, has garnered growing attention in recent years. Despite the rapid advancement of generative models, their evaluation remains largely lagging. Existing methods, whether embedding-based or Multi-modal Large Language Model (MLLM)-based, evaluate alignment with each modal condition in isolation, which contradicts the simultaneous condition alignment objective of multi-modal image generation, leading to poor consistency with human judgments. To address this challenge, we propose \textbf{UFO}, the first \textbf{U}ni\textbf{F}ied framework for \textbf{O}mni-condition alignment simultaneous evaluation. Specifically, UFO introduces a novel Atomized Chain-of-Evaluation paradigm, \emph{i.e.}, it first decomposes omni-condition alignment into a sequential chain of fine-grained, disentangled Atomic Evaluation Units (AEUs), categorizes them into distinct modality-relevance classes, and then employs general or dedicated functional calls for accurate verification of different AEU types. Experimental results demonstrate that UFO achieves the highest correlation with human evaluation preferences, delivering an average improvement of 15.25\%. Furthermore, we present UFO-Bench, a dedicated benchmark designed to holistically evaluate the performance of existing customization models under the diverse mutual interactions of textual and visual conditions.

Image generation models can generate and edit images from text instructions and reference images, but evaluating whether the generated results satisfy all conditions at the same time remains difficult. Existing evaluation methods usually compare text and images separately, which often fails to capture conflicts between the editing instructions and the original visual content. In this work, we propose UFO, a new evaluation framework for subject-driven image generation. UFO breaks complex generation tasks into smaller evaluation units and checks whether generated images both preserve important subject characteristics and follow the requested edits. We also introduce UFO-Bench, a benchmark designed to test image generation models under diverse combinations of text and image conditions. Experiments show that UFO is more consistent with human judgment than existing evaluation methods. Our work provides a more reliable way to evaluate modern image generation systems.