Human accuracy varies by deepfake model and improves with targeted training

From GAN-Based Models to GPT-5: A Taxonomy-Guided Study of Human Performance in Detecting Deepfake Face Images

Computer Vision and Pattern RecognitionCryptography and SecurityHuman-Computer Interaction

Summary

Detecting fake face images made by AI can be hard, and people find it easier or harder depending on how the images were made. The authors studied how well people spot fake faces created by different AI methods, like older GANs and newer models such as GPT-5. They found that training people to recognize specific model mistakes helped them get better at detecting fakes than general training. The improved detection lasted at least five months without making people feel more stressed. This study helps design better ways to train people against deepfake images.

What this means in practice

  • For social media platform security teams: Create targeted user training programs to improve manual detection of deepfake face images from specific AI models used in content moderation.
  • For digital forensics analysts: Develop focused training tools that enhance detection accuracy for various deepfake generation methods encountered in forensic investigations.

Authors

Zhiqing Guo, Lixin Jia, Luning Sun, Xiaowei Chen, Guofeng Yu, Liejun Wang, Gaobo Yang, David Stillwell, Thorsten Holz, Shujun Li

Abstract

Generative models for deepfakes have rapidly evolved from GAN-based architectures to diffusion models and multimodal large language models (MLLMs). However, empirical research on human performance in detecting deepfake face images has not kept pace. It remains unclear if (and how) detection performance varies across generative models and whether model-specific training improves detection more effectively than general-purpose training. In this paper, we propose the \emph{Deepfake Face Artifact Taxonomy} (DFAT), a cross-model taxonomy of human-recognizable visual artifacts in deepfake face images, based on which we build both general-purpose and model-specific training materials. Through an online experiment ($n = 524$), we examined human performance in detecting deepfake face images and evaluated training efficacy across five representative generative models: three GAN-based, one diffusion-based, and one MLLM (GPT-5). Our results show that participants' detection accuracy was strongly model-dependent and that images generated by more recent and advanced models were not necessarily harder to detect. After training, participants achieved significant performance gains across most conditions, and their improvement varied by model. Notably, model-specific training generally yielded greater numerical gains than general-purpose training. In a five-month follow-up test ($n = 342$), we observed no significant decline in detection accuracy and no significant increase in perceived stress. Overall, our findings provide actionable insights for designing effective, sustainable, and targeted training programs to counter the growing threat posed by deepfakes.