A data-driven method for automated facial aesthetic editing that replaces rule-based beauty filters. A real facial image is inverted into the latent space of a pre-trained GAN, and its latent code is iteratively moved along the gradient of a separately trained facial beauty regression network, so the GAN synthesizes a more attractive rendering of the same face. Because the guidance signal is a learned continuous beauty score rather than hand-crafted feature rules, the approach captures holistic patterns of attractiveness directly from data. It combines GAN inversion, gradient-guided latent optimization, and deep perceptual-score regression into a single aesthetic-enhancement pipeline.
Traditional 'beauty filters' enhance facial images with rule-based transformations grounded in predefined attractiveness cues (proportionality, eye size, skin smoothness). This paper proposes a data-…