A low-light illumination is always a problem for some face detection models. However, we can solve this problem by applying a low-light image enhancement on the face detection model. This research evaluates four models of low-light image enhancement, which are MirNet, Adaptive Gamma Correction, RetinexNet, and Retinex. These four models are evaluated using LOL dataset by PSNR and SSIM. The evaluation results are applied on RetinaFace model as face detection and tested using Dark Face dataset. The final model on our proposed method is a combination of both Retinex and RetinaFace face detection models. The result of combined model between Retinex and RetinaFace are outperforms other combined methods. The combine method between low-light image enhancement Retinex model and face detection Retinaface model achieved a mean average precision (mAP) of 0.43%. While, without applying low-light image enhancement model on face detection model, RetinaFace only yielded 0.27% on mAP.