Veritas: Generalizable Deepfake Detection via Pattern-Aware Reasoning

2025-08-28

AI Summary

Existing deepfake detection methods achieve strong performance on academic benchmarks but suffer from poor generalization due to narrow training data distributions and low-quality test images, limiting their applicability in real-world industrial settings. To address this, we propose an open-domain generalizable detection framework. First, we introduce a pattern-aware reasoning mechanism that integrates human-inspired “planning” and “self-reflection” paradigms. Second, we design a two-stage training pipeline synergizing multimodal large language models, hierarchical evaluation protocols, chain-of-thought reasoning, and self-supervised learning—enabling robust detection across diverse generative models, synthesis techniques, and unseen domains. Evaluated on the HydraFake benchmark, our method significantly outperforms state-of-the-art approaches, particularly demonstrating superior generalization to previously unseen forgery techniques and unknown data domains. Moreover, its decision-making process is inherently transparent and interpretable.

Technology Category

• Computer Vision: Large Vision Models
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Deepfake detection remains a formidable challenge due to the complex and evolving nature of fake content in real-world scenarios. However, existing academic benchmarks suffer from severe discrepancies from industrial practice, typically featuring homogeneous training sources and low-quality testing images, which hinder the practical deployments of current detectors. To mitigate this gap, we introduce HydraFake, a dataset that simulates real-world challenges with hierarchical generalization testing. Specifically, HydraFake involves diversified deepfake techniques and in-the-wild forgeries, along with rigorous training and evaluation protocol, covering unseen model architectures, emerging forgery techniques and novel data domains. Building on this resource, we propose Veritas, a multi-modal large language model (MLLM) based deepfake detector. Different from vanilla chain-of-thought (CoT), we introduce pattern-aware reasoning that involves critical reasoning patterns such as "planning" and "self-reflection" to emulate human forensic process. We further propose a two-stage training pipeline to seamlessly internalize such deepfake reasoning capacities into current MLLMs. Experiments on HydraFake dataset reveal that although previous detectors show great generalization on cross-model scenarios, they fall short on unseen forgeries and data domains. Our Veritas achieves significant gains across different OOD scenarios, and is capable of delivering transparent and faithful detection outputs.

Problem

• Addressing generalization challenges in real-world deepfake detection
• Mitigating discrepancies between academic benchmarks and industrial practice
• Detecting unseen forgery techniques and novel data domains

Innovation

• Pattern-aware reasoning with planning and self-reflection
• Two-stage training pipeline for MLLM integration
• Multi-modal large language model based detector