A Comprehensive Survey on AI-Generated Image and Deepfake Detection: From Generative Adversarial Networks to Diffusion Models

The democratization of deep generative algorithms has fundamentally disrupted digital trust, making the creation of hyper-realistic, AI-synthesized media frictionless and accessible to the masses. This comprehensive survey has chronicled the evolutionary trajectory of deepfakes, transitioning from rudimentary autoencoder-based face swaps to highly sophisticated architectures powered by Generative Adversarial Networks (GANs) and, most recently, Text-to-Image Diffusion Models. As generation techniques have grown exponentially in their ability to hallucinate photorealistic pixels, the necessity for robust, automated forensic countermeasures has never been more urgent.   We have detailed the extensive taxonomy of detection methodologies developed to combat this threat. Early detectors relied on spatial-domain CNNs to identify pixel-level artifacts, achieving near-perfect accuracy on controlled, in-distribution datasets like FaceForensics++. However, recognizing the catastrophic generalization failure of these vanilla classifiers, researchers innovated deeply. Modern detection paradigms now target intrinsic manipulation markers—such as blending boundaries via Face X-ray and Self-Blended Images (SBIs)—or exploit severe, periodic upsampling artifacts hiding in the Fourier spectrum utilizing frameworks like-Net. Additionally, leveraging biological signals, such as rPPG and temporal audio-visual dissonance, has provided resilient defenses against sequence-based deepfakes.   Despite these advancements, the adversarial battleground remains fraught with immense challenges. Forensic models continue to struggle against aggressive real-world image degradations (such as H.264 compression and resizing), adversarial evasion techniques designed to erase spectral fingerprints, and the monumental paradigm shift toward Diffusion Models, which bypass the traditional upsampling artifacts that GAN-trained detectors rely upon. To navigate these challenges, the continuous curation of massive, ethically sourced, third-generation benchmark datasets like the DFDC and GenImage is imperative to evaluate detectors under rigorous, "in-the-wild" conditions. Ultimately, winning the war against visual disinformation will require continuous algorithmic innovation, multi-modal analysis, and proactive provenance frameworks to ensure the long-term integrity of digital media.  

[1] Y. Li and S. Lyu, "Exposing deepfake videos by detecting face warping artifacts," in Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2019, pp. 1831–1839. [2] F. Juefei-Xu, R. Wang, Y. Huang, Q. Guo, L. Ma, and Y. Liu, "Countering malicious deepfakes: Survey, battleground, and horizon," arXiv preprint arXiv:2103.00218, 2022. [3] R. Corvi, D. Cozzolino, G. Zingarini, G. Poggi, K. Nagano, and L. Verdoliva, "On the detection of synthetic images generated by diffusion models," arXiv preprint arXiv:2211.00680, 2022. [4] M. Zhu, H. Chen, Q. Yan, X. Huang, G. Lin, W. Li, Z. Tu, H. Hu, J. Hu, and Y. Wang, "GenImage: A million-scale benchmark for detecting AI-generated image," in Advances in Neural Information Processing Systems (NeurIPS), 2023. [5] U. Ojha, Y. Li, and Y. J. Lee, "Towards universal fake image detectors that generalize across generative models," in Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023, pp. 24480–24489. [6] K. Shiohara and T. Yamasaki, "Detecting deepfakes with self-blended images," in Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 18720–18729. [7] H. Zhao, W. Zhou, D. Chen, T. Wei, W. Zhang, and N. Yu, "Multi-attentional deepfake detection," in Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021, pp. 2185–2194. [8] R. Tolosana, R. Vera-Rodriguez, J. Fierrez, A. Morales, and J. Ortega-Garcia, "Deepfakes and beyond: A survey of face manipulation and fake detection," Information Fusion, vol. 64, pp. 131–148, 2020. [9] S.-Y. Wang, O. Wang, R. Zhang, A. Owens, and A. A. Efros, "CNN-generated images are surprisingly easy to spot... for now," in Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020, pp. 8695–8704. [10] Y. Qian, G. Yin, L. Sheng, Z. Chen, and J. Shao, "Thinking in frequency: Face forgery detection by mining frequency-aware clues," in European Conference on Computer Vision (ECCV), 2020, pp. 86–103.