AesCanvas: Evaluating Multimodal Aesthetic Critique and Contextual Suitability
August 28, 2026
AesCanvas introduces CritiqueCanvas, containing 519,136 instruction-response pairs for long-form aesthetic critique, and ContextCanvas, which evaluates suitability across 301 expert-reviewed use scenarios. The benchmark identifies a performance gap between general MLLMs and specialized models in aligning aesthetic quality with specific cultural or domain contexts.
HOW THIS AFFECTS YOU
●
researcherThis provides a large-scale dataset to move MLLM aesthetic assessment from scalar scores to interpretable critiques.
●
designerYou can better evaluate whether generative models understand the nuance of context-specific visual styles.