To address the scarcity of paired data and weak facial feature correlations in low-light face image enhancement, this study proposes a semi-supervised algorithm guided by knowledge distillation and 3D priors. First, a 3D prior extraction network based on a diffusion model is constructed to generate high-quality 3D facial priors in a tri-plane feature space, providing stable structural constraints and detailed guidance for subsequent enhancement. Then, a semi-supervised image enhancement network based on knowledge distillation is designed to jointly leverage paired and unpaired data for low-light face enhancement. The proposed 3D prior feature fusion module enhances the model’s perception of facial spatial structures, thus generating enhanced results with more natural geometry and realistic details. Meanwhile, the unsupervised knowledge distillation strategy combines intra-network and inter-network consistency, effectively enhancing the model’s generalization capability. Experimental results show that the proposed method achieves excellent performance on both synthetic and real low-light face datasets. It significantly improves image brightness and visual realism.