Makeup Transfer aims to apply a given makeup style (extracted from an image called reference) onto a non-makeup face (source) while preserving identity attributes. Makeup transfer is treated as two distinct subtasks : transferring makeup from the reference face to the source face, and removing makeup from the reference face. Recent methods have considered makeup transfer as a color copying task. However, this assumption led to the generation of goodlooking makeup on realistic images instead of realistic makeup. As a result, even though the generated makeup may appear realistic, it is far from the true render of the reference makeup on the skin of the source face. In this paper, the proposed approach aims to preserve the source face’s skin tone information. In addition, to prevent our model from entangling the makeup and makeup removal processes, the model separates the makeup removal and makeup addition tasks. The quantitative results show that we outperform state-of-the-art makeup transfer architectures for both foundation accuracy and consistency. These results are further confirmed by a crowd-sourced user study.