In recent years, machine learning has made significant strides in generating photorealistic facial images. Techniques such as image-to-image transformation have been widely applied across various fields, including data augmentation, entertainment, virtual reality, and the creation of synthetic media like deepfakes. In this talk, I will review key machine learning techniques, including Markov Random Fields (MRF), Principal Component Analysis (PCA), and Generative Adversarial Networks (GANs), for face hallucination. I will also present our recent work on cross- domain heterogeneous face synthesis, with a particular emphasis on facial expression modeling and its applications. Specifically, I will explain how we model facial geometry, textures, and expression styles, and demonstrate how these models enable the generation of realistic images for previously unseen subjects. Finally, I will discuss how this generated data can improve the performance of facial recognition systems.