<p>Multimodal fake news detection is a technique that utilizes multiple data modalities to identify and filter out false or misleading news. Existing approaches to multimodal fake news detection primarily focus on improving the representation of textual and visual content using neural networks. However, these approaches often overlook the valuable potential of hidden information and rely on single decisions, which may lead to inaccuracies when the information is incomplete or ambiguous. To address these limitations, this paper introduces FND-EKA, a hierarchical conditional multimodal fake news detection framework that is augmented with external knowledge, tailored to effectively identify fake news within combined text–image data. First, FND-EKA leverages a knowledge base to extract external knowledge, which is then used to enrich the original textual content and support more accurate verification of news veracity. Secondly, FND-EKA trains a label-supervised information alignment network that projects text and image features into a unified space to achieve aligned text–image representations. Finally, FND-EKA utilizes a hierarchical conditional logic strategy to derive the ultimate classification result. This strategy is proficient in conditionally integrating textual and image data, thereby leveraging additional implicit information. To assess the effectiveness of FND-EKA, we conducted comparative experiments on two popular datasets: Twitter and Weibo. The results indicate that our approach achieves outstanding performance.</p>