FaceForge: Personalized Facial Enhancement with Class-guided Noise Reduction

Abstract

Personalizing digital faces is a fundamental task in various fields such as virtual try-on, digital avatars, and face image editing. However, achieving realistic and editable face personalization remains a challenge due to the inherent noise and variability of face images. This study presents FaceForge, a new approach to editable face personalization using class-driven noise suppression regulation. FaceForge uses a diffusion-based generative model trained with class-specific information to enable high-quality facial manipulation while preserving identity and expression details. The proposed method effectively suppresses noise and artifacts in face images, allowing users to personalize faces with unprecedented fidelity and control. Experimental results demonstrate the superior performance of FaceForge compared to state-of-the-art methods in terms of image quality, editability, and robustness to input variations. In addition, we demonstrate the practical implications of FaceForge in real-world applications such as virtual testing systems and the creation of digital avatars. Overall, FaceForge represents a significant advance in the field of facial personalization, offering users powerful tool to easily create personalized and distinctive digital faces