Following the previous chapter on 2D facial attractiveness prediction, this chapter focuses on 2D beauty enhancement, namely editing faces to increase attractiveness while preserving identity and realism. Face beautification is an emerging application widely used in the entertainment industry. We first review traditional enhancement methods in the geometric and photometric domains, including landmark-driven shape adjustment, local smoothing, and skin tone refinement. We then survey modern deep-learning-based approaches, particularly GAN-based methods, and discuss their strengths and limitations in semantic control, geometric fidelity, and background preservation. Most of these treat facial style codes as a black box, lack semantic interpretability, and often overlook geometric cues or alter background content. To address these issues, we introduce ISFB-GAN, an interpretable semantic framework that factorizes facial style into geometry and appearance via a dual-branch encoder. An attention-guided generator focuses edits on beauty-related regions while preserving background content. Training employs a hybrid objective combining geometric, orthogonality, segmentation, style reconstruction, and perceptual losses to improve disentanglement and visual fidelity. Experiments on CelebAMask-HQ and SCUT-FBP5500 demonstrate superior texture refinement and subtle geometric enhancement while maintaining identity and background consistency, surpassing representative state-of-the-art methods in facial attractiveness enhancement.