Detecting Localized Deepfakes: How Well Do Synthetic Image Detectors Handle Inpainting?

The rapid progress of generative AI has enabled highly realistic image manipulations, including inpainting and region-level editing. These approaches preserve most of the original visual context and are increasingly exploited in cybersecurity-relevant threat scenarios. While numerous detectors have been proposed for identifying fully synthetic images, their ability to generalize to localized manipulations remains insufficiently characterized. This work presents a systematic evaluation of state-of-the-art detectors, originally trained for the deepfake detection on fully synthetic images, when applied to a distinct challenge: localized inpainting detection. The study leverages multiple datasets spanning diverse generators, mask sizes, and inpainting techniques. Our experiments show that models trained on a large set of generators exhibit partial transferability to inpainting-based edits and can reliably detect medium- and large-area manipulations or regeneration-style inpainting, outperforming many existing ad hoc detection approaches.

Key Contributions

Systematic evaluation protocol for state-of-the-art synthetic image detectors applied to localized inpainting detection across diverse generators, mask sizes, and manipulation types
Analysis of transferability gaps between fully synthetic image detection and localized deepfake detection, identifying conditions (large masks, regeneration-style inpainting) where transfer succeeds
Actionable insights on which detector design factors (training generator diversity) most influence generalization to inpainting-based manipulations

🛡️ Threat Analysis

Output Integrity Attack

Directly addresses AI-generated content detection — specifically evaluating whether deepfake detectors generalize from fully synthetic images to localized inpainting manipulations, which is an output integrity and content authenticity problem.