We propose a working framework, FaceLit, that is able to generate a 3D face that can be rendered in different user-defined lighting conditions and scenes, completely learned from 2D images in the wild without any manual annotation. Unlike current businesses that require careful setup for pickup or human labor, we rely on turnkey and lighting estimates. Using these estimates, we integrate the Phong reflection model into a neural volume rendering framework. Our model learns to create the shape and material properties of a face so that, when viewed according to the natural statistics of posture and illumination, it produces realistic images of the face with multiple 3D rendering and lighting consistency. Our method enables the creation of realistic faces with clear lighting and rendering controls on multiple datasets – FFHQ, MetFaces, and CelebA-HQ. We show the latest real-time images among 3D-aware GANs on the FFHQ dataset that achieved a FID score of 3.5.