Due to the rapid growth in computational capabilities and the misuse of technology, exemplified by the production of lifelike face-swapped deep fakes for nefarious purposes such as sowing political discord, fabricating terrorism incidents, disseminating revenge pornography, and engaging in blackmail, this manuscript introduces an innovative deep learning methodology. The methodology, named Deep Truth Guardian (DTG), is designed to effectively discern AI-generated counterfeit videos from authentic ones, with a specific focus on identifying deep fakes involving scene replacement and re-enactment. We employed a ResNet convolutional neural network to extract intricate frame-level features. These extracted features serve as input for training a long short-term memory (LSTM)-based recurrent neural network (RNN), enabling the classification of videos based on the presence of manipulation, thereby distinguishing genuine videos from deepfakes. The proposed method, DTG, utilized diverse and sizable datasets such as Face-Forensic++, the deepfake Detection Challenge, and Celeb-DF. DTG makes a significant contribution to safeguarding the authenticity of visual media in an era characterized by sophisticated AI-generated manipulations. This proposed system, Deep Truth Guardian, sets a new standard for deepfake detection with its high accuracy and adaptability, representing a significant step toward safeguarding the integrity of digital content in an era marked by advanced AI-driven manipulations.