Visual Aesthetic Benchmark: Can Frontier Models Judge Beauty?

2026-05-12

AI Summary

This work addresses the limitation of existing visual aesthetic assessment methods, which predominantly rely on absolute scoring of individual images and fail to capture human relative preferences. The authors propose modeling aesthetic judgment as a comparative selection task among candidate images sharing the same theme, introducing a novel set-wise comparative evaluation paradigm grounded in expert consensus. They construct VAB, a visual aesthetics benchmark comprising 400 tasks and 1,195 images. Experiments reveal that even the strongest current multimodal models correctly identify both the best and worst images in only 26.5% of tasks, substantially below the human expert performance of 68.9%. Notably, fine-tuning a 35B-parameter model with a modest amount of expert data achieves performance approaching that of a 397B-parameter model, demonstrating the efficacy and promise of the proposed paradigm.

Abstract

Multimodal large language models (MLLMs) are now routinely deployed for visual understanding, generation, and curation. A substantial fraction of these applications require an explicit aesthetic judgment. Most existing solutions reduce this judgment to predicting a scalar score for a single image. We first ask whether such scores faithfully capture comparative preference: in a controlled study with eight expert annotators, score-derived rankings align poorly with the same annotators' direct comparisons, while direct ranking yields substantially higher inter-annotator agreement on best- and worst-image labels. Motivated by this finding, we introduce the Visual Aesthetic Benchmark (VAB), which casts aesthetic evaluation as comparative selection over candidate sets with matched subject matter. VAB contains 400 tasks and 1,195 images across fine art, photography, and illustration, with labels derived from the consensus of 10 independent expert judges per task. Evaluating 20 frontier MLLMs and six dedicated visual-quality reward models, we find that the strongest system identifies both the best and the worst image correctly across three random permutations of the candidate order in only 26.5% of tasks, far below the 68.9% achieved by human experts. Fine-tuning a 35B-parameter model on 2,000 expert examples brings its accuracy close to that of a 397B-parameter open-weight model, suggesting that the comparative signal in VAB is transferable. Together, these results expose a clear and measurable gap between current multimodal models and expert aesthetic judgment, and VAB provides the first set-based, expert-grounded testbed on which that gap can be tracked and closed.

Problem

• aesthetic judgment
• multimodal large language models
• visual understanding
• comparative preference
• expert annotation

Innovation

• comparative aesthetic judgment
• Visual Aesthetic Benchmark
• multimodal large language models
• expert-grounded evaluation
• aesthetic preference ranking