In recent years, deep learning has revolutionized 3D face modeling and reconstruction, allowing us to create highly detailed models of facial shapes from everyday images and videos. However, capturing the full range of facial expressions, especially subtle or less common ones, remains a challenge. In this talk, we will explore how deep learning is helping to overcome these obstacles, focusing on techniques that improve the realism and accuracy of facial expressions in 3D models. We will see how specific methods, including self-supervision, enhance the reconstruction of expressions, such as those linked to speech and emotions. By examining these approaches, we’ll gain insights into how deep learning techniques can improve the precision and applicability of 3D face reconstruction in various contexts.