In vivo tau-positron emission tomography (PET) is an important method to determine the stage of Alzheimer’s disease (AD), guiding treatment decisions and influencing prognosis. However, tau-PET is expensive, not widely available and exposes patients to ionizing radiation, which poses a carcinogenic risk. To address this issue, we propose to use a generative adversarial network (GAN) to synthesize tau-PET brain images from T1-weighted brain images. The build GAN model is based on a 3D conditional generative adversarial network and employs multiscale discriminators and feature matching to ensure realistic details. We will train the network on a set of public medical datasets to learn a network that can map the representation across different cohorts. This work can potentially increase the clinical value of T1-weighted images, facilitating multi-modal diagnosis for AD, as well as reducing PET imaging costs and associated radiation exposure.