MakeupMirror: Improving Facial Attribute Preservation in Diffusion Models for Makeup Transfer

MakeupMirror is a newly proposed diffusion-based model for makeup transfer that enhances facial attribute preservation, addressing limitations in identity and skin color retention seen in prior models like Stable-Makeup. Key innovations include facial geometry conditioning with ControlNets, region-specific makeup application, skin tone modulation, and a Levenberg-Marquardt Langevin sampler for faster inference, achieving a 60% improvement in facial recognition similarity and a 50% reduction in skin tone differences. This model is significant for practitioners as it enables more realistic virtual try-on experiences in makeup shopping, enhancing user satisfaction and accuracy in augmented reality applications.