This paper explores the detection of deepfake content, particularly focusing on multiclass deepfake images created using Generative Adversarial Networks (GANs) and Stable Diffusion-based methods. The primary objective is to assess the effectiveness of Vision Transformers (ViTs) in detecting deepfake images compared to traditional Convolutional Neural Network (CNN)-based models. The study introduces a novel approach by framing the deepfake detection problem as a multiclass task, addressing the challenges posed by advanced techniques like Stable Diffusion and StyleGAN2. - **Objective**: To evaluate the viability of ViTs in detecting multiclass deepfake images. - **Novelty**: Approaching deepfake detection as a multiclass task and leveraging ViTs for global feature extraction. - **Results and Conclusion**: The proposed method achieves high accuracy (99.90%) on a multiclass-prepared dataset, outperforming state-of-the-art CNN models like ResNet-50 and VGG-16. - Deep learning - Image processing - CNN - Deepfake identification - Artificial intelligence - Stable Diffusion - StyleGAN2 - Vision transformer - Global feature extraction - Patches The paper begins by discussing the rapid advancements in facial manipulation technologies and the need for robust detection methods. It then outlines the primary goal of the study, which is to assess the effectiveness of ViTs in detecting multiclass deepfake images. The novelty of the research lies in its approach to the deepfake detection problem as a multiclass task and the use of ViTs for global feature extraction. The results section highlights the high accuracy and precision of the proposed method, surpassing state-of-the-art CNN models. The discussion section interprets the findings and outlines future implications for multimedia forensics. The conclusion emphasizes the significance of the proposed method in combating deepfake threats and promoting digital integrity.This paper explores the detection of deepfake content, particularly focusing on multiclass deepfake images created using Generative Adversarial Networks (GANs) and Stable Diffusion-based methods. The primary objective is to assess the effectiveness of Vision Transformers (ViTs) in detecting deepfake images compared to traditional Convolutional Neural Network (CNN)-based models. The study introduces a novel approach by framing the deepfake detection problem as a multiclass task, addressing the challenges posed by advanced techniques like Stable Diffusion and StyleGAN2. - **Objective**: To evaluate the viability of ViTs in detecting multiclass deepfake images. - **Novelty**: Approaching deepfake detection as a multiclass task and leveraging ViTs for global feature extraction. - **Results and Conclusion**: The proposed method achieves high accuracy (99.90%) on a multiclass-prepared dataset, outperforming state-of-the-art CNN models like ResNet-50 and VGG-16. - Deep learning - Image processing - CNN - Deepfake identification - Artificial intelligence - Stable Diffusion - StyleGAN2 - Vision transformer - Global feature extraction - Patches The paper begins by discussing the rapid advancements in facial manipulation technologies and the need for robust detection methods. It then outlines the primary goal of the study, which is to assess the effectiveness of ViTs in detecting multiclass deepfake images. The novelty of the research lies in its approach to the deepfake detection problem as a multiclass task and the use of ViTs for global feature extraction. The results section highlights the high accuracy and precision of the proposed method, surpassing state-of-the-art CNN models. The discussion section interprets the findings and outlines future implications for multimedia forensics. The conclusion emphasizes the significance of the proposed method in combating deepfake threats and promoting digital integrity.