Gen Defend-STF Net: Deepfake-Resistant Biometric Security for Trustworthy Digital Finance, Commerce and Business Management in the GenAI Era

The rapid advancement of Generative AI (GenAI) has enabled the creation of highly realistic deepfake images and videos, posing critical threats to biometric authentication systems. Deepfakes generated through GANs, diffusion models, and neural rendering techniques can mimic facial appearance, motion, and texture with unprecedented accuracy, making conventional liveness detection and recognition methods increasingly vulnerable. To address this growing challenge, this paper proposes GenDefend-STFNet, a novel Multi-Modal Spatio-Temporal Fusion Network designed to provide trustworthy and robust defence against GenAI-powered deepfake attacks. The framework integrates four complementary feature streams: (i) spatial texture features from RGB input, (ii) temporal motion irregularities detected through a lightweight transformer encoder, (iii) frequency-domain artefacts that reveal deepfake synthesis patterns, and (iv) depth-based structural cues for identifying 3D inconsistencies. A cross-attention fusion module combines these modalities to capture both fine-grained and holistic anomalies associated with synthetic facial content. Experimental evaluations on benchmark deepfake datasets and custom GenAI-generated samples demonstrate that GenDefend-STFNet significantly outperforms state-of-the-art anti-spoofing and deepfake detectors in accuracy, generalisation, and robustness. The proposed system provides a scalable, explainable, and future-ready biometric security framework suitable for deployment in high-risk identity verification environments.