Deepfakes, advanced Artificial Intelligence (AI) generated media, pose a significant threat to digital information integrity, raising concerns regarding online trust and security. These highly realistic synthetic artifacts are produced using increasingly complex generative techniques. Traditional deepfake detection methods primarily focus on identifying manipulations generated by a single technique. However, they falter when confronted with the intricacies of “deepfake phylogeny”—a scenario involving iterative applications of multiple deepfake generation techniques, resulting in layered manipulations that obscure the content’s origin and evolutionary process. This research presents two hybrid deep learning models aimed at tackling the challenge of identifying the sequence in which these generative techniques are applied. The first model combines ResNeXt and Swin Transformer, effectively capturing both fine-grained and global patterns in manipulated videos. The second model integrates ResNeXt with Vision Transformer, utilizing convolutional and attention mechanisms to enhance detection of the complex order of operations involved in deepfake creation. Both models address an 18-class classification task, predicting the sequence of techniques such as FSGAN, FaceSwap, and FaceShifter. This research addresses the escalating prevalence of deepfakes in the Indian digital landscape and the pressing need for advanced detection mechanisms, contributing to the development of resilient algorithms and enhancing the capacity for forensic analysis in digital environments.