Towards High Fidelity Face Swapping: A Comprehensive Survey and New Benchmark

Authors: Qi Li, Weining Wang, Shuangjun Du, Bo Peng, Jing Dong, Kun Wang, Zhenan Sun, Ming-Hsuan Yang

Published: 2026-04-27 03:00:00+00:00

AI Summary

This paper provides a comprehensive survey and benchmark for face swapping, organizing existing methods into five major paradigms. It introduces CASIA FaceSwapping, a high-quality video benchmark dataset with balanced demographic distributions and explicit attribute variations. Standardized protocols are established to evaluate the robustness and generalization ability of different face swapping methods, yielding new insights into their performance characteristics and limitations.

Abstract

Face swapping has witnessed significant progress in recent years, largely driven by advances in deep generative models such as GANs and diffusion models.Despite these advances, existing methods remain fragmented across different paradigms, and their evaluation is highly inconsistent due to the lack of standardized datasets and protocols. Moreover, prior surveys primarily focus on broader deepfake generation or detection, leaving face swapping insufficiently studied as a standalone problem. In this paper, we present a comprehensive survey and benchmark for face swapping. We provide a structured review of existing methods, organizing them into five major paradigms and systematically analyzing their design principles, strengths, and limitations. To enable fair and controlled evaluation, we introduce CASIA FaceSwapping, a high-quality benchmark with balanced demographic distributions and explicit attribute variations, and establish standardized protocols to assess the robustness of different face swapping methods. Extensive experiments on representative approaches yield new insights into the performance characteristics and limitations of current techniques. Overall, our work provides a unified perspective and a principled evaluation framework to facilitate the development of more robust and controllable face swapping methods. More results can be found at https://github.com/CASIA-NLPRAI/face-swapping-survey.

Key findings

InsightFace consistently demonstrated the strongest identity preservation across all evaluation protocols, while SimSwap excelled in preserving pose and expression attributes. Most methods experienced a performance drop under cross-ethnicity and cross-attribute conditions, indicating challenges in cross-domain generalization and robustness. The study highlights that identity preservation and attribute consistency are competing objectives, with methods like FaceAdapter achieving a practical balance between realism and identity fidelity.

Approach

The authors conducted a comprehensive survey by classifying existing face swapping methods into five paradigms: 3DMM-based, autoencoder-based, GAN-based, StyleGAN-based, and diffusion-based. They introduced a new large-scale video benchmark dataset, CASIA FaceSwapping, with 4K resolution, balanced demographics, and controlled attribute variations. Furthermore, three standardized evaluation protocols (Normal, Cross-ethnicity, Cross-attribute) were established to rigorously assess face swapping algorithms.

Datasets

CASIA FaceSwapping, UADFV, Deepfake TIMIT, FaceForensics++, Celeb-DF, DeeperForensics, DFDC

Model(s)

UNKNOWN

Author countries

China, Singapore, USA, Korea

← Previous