ML Security PapersCLUE: Leveraging Low-Rank Adaptation to Capture Latent Uncovered Evidence for Image Forgery Localization
Youqi Wang 1, Shunquan Tan 2, Rongxuan Peng 1, Bin Li 2, Jiwu Huang 2
Published on arXiv
2508.07413
Output Integrity Attack
OWASP ML Top 10 — ML09
Key Finding
CLUE achieves state-of-the-art generalization and robustness across multiple public forgery localization benchmarks, significantly outperforming prior methods on both traditional and AI-generated forgeries
CLUE (Capture Latent Uncovered Evidence)
Novel technique introduced
The increasing accessibility of image editing tools and generative AI has led to a proliferation of visually convincing forgeries, compromising the authenticity of digital media. In this paper, in addition to leveraging distortions from conventional forgeries, we repurpose the mechanism of a state-of-the-art (SOTA) text-to-image synthesis model by exploiting its internal generative process, turning it into a high-fidelity forgery localization tool. To this end, we propose CLUE (Capture Latent Uncovered Evidence), a framework that employs Low- Rank Adaptation (LoRA) to parameter-efficiently reconfigure Stable Diffusion 3 (SD3) as a forensic feature extractor. Our approach begins with the strategic use of SD3's Rectified Flow (RF) mechanism to inject noise at varying intensities into the latent representation, thereby steering the LoRAtuned denoising process to amplify subtle statistical inconsistencies indicative of a forgery. To complement the latent analysis with high-level semantic context and precise spatial details, our method incorporates contextual features from the image encoder of the Segment Anything Model (SAM), which is parameter-efficiently adapted to better trace the boundaries of forged regions. Extensive evaluations demonstrate CLUE's SOTA generalization performance, significantly outperforming prior methods. Furthermore, CLUE shows superior robustness against common post-processing attacks and Online Social Networks (OSNs). Code is publicly available at https://github.com/SZAISEC/CLUE.
Key Contributions
- First framework to repurpose Stable Diffusion 3's internal generative process (via LoRA parameter-efficient adaptation) as a forensic feature extractor for image forgery localization
- Demonstrates that SD3's Rectified Flow noise mechanism, guided by LoRA fine-tuning, amplifies subtle statistical inconsistencies in forged image regions across multiple noise levels
- Integrates LoRA-adapted SAM image encoder for spatial-semantic context, achieving SOTA generalization against both traditional forgeries (copy-move, splicing) and AI-generated forgeries with robustness to post-processing and OSNs
🛡️ Threat Analysis
Proposes a novel AI-generated content detection and localization framework — detecting tampered/forged image regions including those produced by generative AI models. This is a forensic technique for output/content integrity, fitting squarely within ML09's scope of AI-generated content detection and content authenticity.