VisualDeltas: Learning Preferences from Visual Quality Perturbations

ArXi:2603.07272v1 Announce Type: new We present VisualDeltas, a lightweight preference-learning framework that extracts supervision from visual quality variations in multimodal data. By leveraging the systematic impact of image quality on visual perception and reasoning, VisualDeltas induces informative preference signals without relying on human annotations or external teachers. The framework s both label-free and label-based regimes, enabling flexible use of available supervision when present.