Evidence Packing for Cross-Domain Image Deepfake Detection with LVLMs

TL;DR

This paper introduces SCEP, a training-free framework that enhances deepfake detection by selecting and reasoning over suspicious image patches using LVLMs, avoiding costly fine-tuning.

Contribution

SCEP is a novel evidence-driven approach that leverages LVLMs for cross-domain image deepfake detection without requiring model fine-tuning.

Findings

SCEP outperforms strong baselines on multiple benchmarks.

It effectively detects diverse and evolving deepfakes.

The method reduces reliance on costly model fine-tuning.

Abstract

Image Deepfake Detection (IDD) separates manipulated images from authentic ones by spotting artifacts of synthesis or tampering. Although large vision-language models (LVLMs) offer strong image understanding, adapting them to IDD often demands costly fine-tuning and generalizes poorly to diverse, evolving manipulations. We propose the Semantic Consistent Evidence Pack (SCEP), a training-free LVLM framework that replaces whole-image inference with evidence-driven reasoning. SCEP mines a compact set of suspicious patch tokens that best reveal manipulation cues. It uses the vision encoder's CLS token as a global reference, clusters patch features into coherent groups, and scores patches with a fused metric combining CLS-guided semantic mismatch with frequency-and noise-based anomalies. To cover dispersed traces and avoid redundancy, SCEP samples a few high-confidence patches per cluster and applies grid-based NMS, producing an evidence pack that conditions a frozen LVLM for prediction. Experiments on diverse benchmarks show SCEP outperforms strong baselines without LVLM fine-tuning.