Published Paper Details:

Face swapping has become an important area of research in computer vision. It is driven by uses in entertainment, privacy, virtual reality, and digital forensics. This review summarizes progress from ten recent studies focusing on GAN-based, diffusion-based, and hybrid models. These studies tackle issues like high fidelity, keeping identities intact, handling occlusions, pose invariance, and scalability. Early GAN-based models such as SimSwap, FaceShifter, and FaceDancer introduced new techniques like identity injection, adaptive embedding fusion, and occlusion-aware refinement. These methods aim to balance transferring the source identity with preserving target attributes. High-resolution techniques like MegaFS used hierarchical latent encoding and StyleGAN2 synthesis for megapixel swaps. In contrast, low-resolution and occlusion-focused methods applied cross-resolution contrastive learning and special inpainting pipelines. Diffusion-based techniques, including DiffSwap, DiffFace, and unified inpainting frameworks, have improved stability, control, and image realism. They achieved this through identity guided sampling, 3D-aware masking, and CLIP feature disentanglement. Together, these advancements show significant development in creating realistic, high-resolution, and attribute-consistent swapped faces. However, there are still challenges with extreme pose variation, detailed preservation, and real-time efficiency. This review discusses the strengths and weaknesses of these approaches and suggests future research directions for creating reliable, ethical, and controllable face swapping systems.