The rapid development of artificial intelligence techniques has led to the emergence of deepfakes. Deepfake is the term specifically used to describe face manipulation produced through deep learning methods. Recent advertisements featuring face manipulation have presented the manipulated object as genuine, despite its creation by an artificial intelligence tool designed for generating deepfakes. Researchers in the field have outlined various methods to detect deepfakes, including the incorporation of varied data types, image alteration, and several neural network techniques, such as convolutional neural networks and ResNet. Current models typically focus on detecting a single object. However, when considering multiple objects, most algorithms have struggled and resulted in low accuracy and lack of interpretability. The Granulated Ensemble Convolutional Neural Network work proposes a new mathematical model to enhance detection accuracy for deepfakes involving multiple objects environment. The word granulated refer to fine grain, segmented and localized analysis of objects in a multi object environment. We evaluate this proposed model against existing deepfake detection techniques such as deepfake assessment frameworks, data augmentation methods, convolutional vision transformers, and generative convolutional vision transformers. The results indicate that the model achieves an accuracy of 94.07%. Experimental results on the DFDC and DF Multi-object datasets demonstrate that the suggested approach achieves an accuracy of 94.07%, which is an 80% improvement over classical models such as Single-Image CNN, GNN, and Reset. Furthermore, compared to other well-established methods, the suggested solution shows an average inference time that is roughly 23% shorter, indicating both increased precision and resource efficiency. These results underscore the model’s effectiveness for practical applications. Its performance was evaluated using the DFDC Face Forensics dataset.