USENIX Security2025Top-tier venue
Chimera: Creating Digitally Signed Fake Photos by Fooling Image Recapture and Deepfake Detectors
Seongbin Park, Alexander Vilesov, Jinghuai Zhang, Hossein Khalili, Yuan Tian, Achuta Kadambi, Nader Sehatbakhsh
Abstract
Deepfake detectors relying on heuristics and machine learning are locked in a perpetual struggle against evolving attacks. In contrast, cryptographic solutions provide strong safeguards against deepfakes by creating hardware-binding digital signatures when capturing (real) images. While effective, they falter when attackers misuse cameras to recapture images of digitally generated fake images from a display or other medium. This vulnerability reduces the security assurance back to the effectiveness of deepfake detectors. The main difference, however, is that a successful attack must now deceive two types of detectors simultaneously: deepfake detectors and detectors specialized for detecting image recaptures. This paper introduces Chimera, an end-to-end attack strategy that crafts cryptographically signed fake images capable of deceiving both deepfake and image recapture detectors. Chimera demonstrates that current adversarial and generative models fail to effectively deceive both detector types or lack generalization across different setups. Chimera addresses this gap by using a hardware-aware adversarial compensator to craft fake images that successfully bypass stateof-the-art detection mechanisms. The key innovation is a GAN-based image generator that accounts for and compensates the physical transformations introduced during the recapture process. Through rigorous testing using commercial off-the-shelf cameras and displays, Chimera proves effective in fooling both types of detectors with a high success rate while having high visual quality (compared to the original real image). Chimera demonstrates the vulnerability of deepfake detectors even when equipped with hardware-based digital signatures. Our successful end-to-end attack on state-of-theart detectors shows an urgent need for more robust detection and mitigation strategies. The source code is available at https://github.com/ssysarch/Chimera . (the author order does not reflect their extent of contributions). Deepfake Detector Real Sign Real Fake Verify Deepfake Detector Verify Real Real Fake (a) While existing deepfake detection solutions are unable to defend against adaptive attacks, cryptographic signatures can protect the system by creating tamper-proof signatures. Sign Sign Deepfake Detector Verify Recapture Detector State-of-the-Art Our Model Real Fake + + (b) Cryptographic methods are vulnerable to screenshot deepfake attacks. Successful attacks, however, need to fool both deepfake AND recapture detectors. Our scheme, Chimera, is the first method to achieve such a capability.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
1 top-tier paper has cited it since. Know what they found before your reviewers do.
Lune can draw on
the full paper · the top-tier papers that cite it · related work
Free to start. No credit card required.
Lineage
- ViGText: Deepfake Image Detection with Vision-Language Model Explanations and Graph Neural Networks cites An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
- ViGText: Deepfake Image Detection with Vision-Language Model Explanations and Graph Neural Networks cites Learning Transferable Visual Models From Natural Language Supervision
- High-Resolution Image Synthesis with Latent Diffusion Models cites Diffusion Models Beat GANs on Image Synthesis
- High-Resolution Image Synthesis with Latent Diffusion Models cites Denoising Diffusion Probabilistic Models
- Diffusion Models Beat GANs on Image Synthesis cites Learning Transferable Visual Models From Natural Language Supervision
- Diffusion Models Beat GANs on Image Synthesis cites Denoising Diffusion Probabilistic Models
- Learning Transferable Visual Models From Natural Language Supervision cites An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
Cited by top-tier papers1
Builds on18
- Learning Transferable Visual Models From Natural Language SupervisionAlec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh et al.ICML 2021 · 47,906 citations
- Denoising Diffusion Probabilistic ModelsJonathan Ho, Ajay Jain, Pieter AbbeelNeurIPS 2020 · 35,902 citations
- An Image is Worth 16x16 Words: Transformers for Image Recognition at ScaleAlexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn et al.ICLR 2021 · 21,477 citations
- Diffusion Models Beat GANs on Image SynthesisPrafulla Dhariwal, Alexander Quinn NicholNeurIPS 2021 · 13,211 citations
- High-Resolution Image Synthesis with Latent Diffusion ModelsRobin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser et al.CVPR 2022 · 13,123 citations
Related papers
- AVA: Inconspicuous Attribute Variation-based Adversarial Attack bypassing DeepFake DetectionXiangtao Meng, Li Wang, Shanqing Guo, Lei Ju et al.S&P 2024 · 17 citations
- CMUA-Watermark: A Cross-Model Universal Adversarial Watermark for Combating DeepfakesHao Huang, Yongtao Wang, Zhaoyu Chen, Yuze Zhang et al.AAAI 2022 · 131 citations
- An Analysis of Recent Advances in Deepfake Image Detection in an Evolving Threat LandscapeSifat Muhammad Abdullah, Aravind Cheruvu, Shravya Kanchi, Taejoong Chung et al.S&P 2024 · 42 citations
- SiFMimicEvader: Evading Fake Voice Detection with Adversarial Neural Mimicry AttacksXuan Hai, Xin Liu, Zihao Zhang, Ziyao Yu et al.ACM MM 2025
- Revisiting Adversarial Patches for Designing Camera-Agnostic Attacks against Person DetectionHui Wei, Zhixiang Wang, Kewei Zhang, Jiaqi Hou et al.NeurIPS 2024 · 22 citations
