The widespread dissemination of fake news poses a significant threat to the integrity of information and societal trust. This study introduces an advanced framework for detecting fake news by integrating transformer-based models with multimodal deep learning techniques. The proposed methodology combines text analysis using models like BERT, image processing with convolutional neural networks, and temporal analysis of video data through transformer architectures. By employing attention-based fusion mechanisms, the framework effectively merges features across modalities to identify patterns indicative of misinformation. Comprehensive preprocessing, feature extraction, and classification processes ensure robust performance. Experiments validate the framework’s efficacy using diverse datasets, achieving high precision and reliability in identifying fake news across multiple formats. This research sets a foundation for developing sophisticated multimodal systems to combat the growing challenges of misinformation.