The rapid advancement of image-to-image translation technology has raised concerns about its potential misuse in image-related criminal activities. In this paper, we are the first to propose a watermarking method specially designed for image-to-image translation in facial editing, named WIT-Net. WIT-Net is capable of generating diverse watermarked translated images from an original facial image. To achieve this, we introduce a novel message encoder that transforms watermarks into a message-matrix, which is efficiently embedded into the translated images. Once the translation process is complete, the translated images are embedded with the watermarks for tracking and marking. Additionally, we propose a restore enhancement module to assist receivers in reconstructing high-quality original images from the translated versions. This module leverages semantic features extracted from pre-restored images and texture features from the translated images to generate the restored images. Extensive experiments validate the effectiveness of our proposed method in producing watermarked translated images and accurately restoring them to their original forms.