Intelligent face-swapping was widely used in the fields of entertainment, movie and television production, and privacy protection, but there were still deficiencies in image quality, realism, and coordination with the original scene. To address this problem, the paper combines the fusion of generative adversarial network (GAN) and auto-encoder (AE) technology based on DeepFaceLab to propose an improved model of intelligent face-swapping, and analyzes the effects of different hardware, different lighting, and different occlusion conditions on the intelligent face-swapping technology. The results show that the model has strong robustness in light change and occlusion environments, and the number of iterations can effectively improve the performance of PSNR (28.5dB) and SSIM (0.91).