Manuscript received October 17, 2025; revised December 10, 2025; accepted April 8, 2026; published November 13, 2026.

Abstract—Deepfakes are becoming more authentic, which lowers confidence in digital media. This paper presents a lightweight and reproducible baseline for video deepfake detection that is tested on the English and Arabic subsets of the PolyGlotFake dataset. The method samples frames from each video and classifies them with a ResNet-18 model, and combines frame predictions to make a clip-level prediction. The proposed method can achieve an accuracy of 97.75% and a macro-Area Under the Curve (AUC) of 0.999, and a Matthews Correlation Coefficient (MCC) of 0.970 on a person-disjoint test split, demonstrating strong class separability and low computational cost. This work offers a grounded basis of practicality in bilingual video deepfake detection, through the provision of an open experimental pipeline and competitive functionality, as well as a basis for future temporal and multimodal extensions.

Keywords—deepfake detection, video forensics, convolutional neural networks, English, Arabic

Cite: Rawad Abdulghafor, Yousuf Al Husaini, Mohammad Abrar, Abdullah Said Al-Aamri, Mohammed Abdulla Salim Al Husaini, and Wasin AlKishri, "Frame-to-clip CNN Baseline for Video Deepfake Detection in English and Arabic," Journal of Image and Graphics, Vol. 14, No. 6, pp. 966-973, 2026.

Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).