DETECTING AI-GENERATED IMPERSONATIONS AND DEEPFAKE MISUSE OF NANO BANANA OUTPUTS

Abstract

This paper addresses the pressing challenge of de- tecting AI-generated impersonations and deepfake misuse stem- ming from Nano Banana, a cutting-edge visual synthesis model. We propose a novel hybrid detection framework that leverages DenseNet121 convolutional architecture combined with frequency- domain signal decomposition and high-order texture feature extraction to capture both spatial and spectral anomalies in- herent in synthetic media. By uniting deep spatial representation learning with frequency-aware analysis, the system effectively discriminates between authentic and Nano Banana-generated facial and object images. The model was trained and validated on a rigorously curated dataset composed of diverse real and synthetic image samples, specifically tailored to reflect real-world impersonation scenarios. Experimental results demonstrate that our approach achieves a precision of 85.87%, significantly surpassing conventional CNN baselines. These find- ings suggest that integrating DenseNet121 with frequency-domain techniques enhances the detection of subtle artifacts that evade purely spatial methods. The framework offers a promising tool for strengthening online identity verification, multimedia forensics, and automated trust evaluation workflows, addressing the growing risks posed by advanced synthetic content. This research contributes a scalable and reliable methodology for safeguarding digital media authenticity in an era of rapidly evolving