<p>The rapid gain of fake news and misinformation on online platforms poses significant societal challenges. Traditional detection methods primarily focus on textual analysis, often failing to account for the contextual influence of accompanying images. To address this limitation, this paper introduces a multimodal weighted attention-based feature fusion approach for fake news detection (FND). A Generative Adversarial Network (GAN) is employed to generate multiple images using the textual content of news articles. These generated images are then compared with the actual image, selecting the most visually similar one as additional contextual information. BERT and pretrained CNN models are utilized to extract textual and visual features, respectively. The integration of weighted attention-based feature fusion effectively captures complex relationships between these modalities. Additionally, GANs contribute to dataset augmentation by simulating evolving fake news patterns, enhancing model diversity and robustness. Extensive evaluations demonstrate that the proposed multimodal framework improves overall detection accuracy by 2.8%, outperforming traditional single-modality as well as multimodal approaches.</p>