The spread of fake news, particularly across social media and digital news platforms, poses a growing challenge due to its influence on public opinion and social stability. This study introduces a novel framework for fake news detection, leveraging Large Language Models (LLMs), specifically GPT-4, with multimodal data fusion and an external fact-checking module. Unlike conventional approaches that rely mainly on text, the proposed framework addresses the multimodal nature of misinformation by integrating textual and visual features captured through Convolutional Neural Networks (CNNs) for images to detect misleading content more effectively. Additionally, the framework incorporates real-time fact-checking by cross-referencing statements with trusted sources, such as Wikipedia and FactCheck.org, reducing false positives and model hallucinations. This approach beats current methods, offering a more thorough solution to the challenging problem of disinformation detection, according to extensive testing on benchmark datasets. Future work will explore multilingual and real-time applications to enhance its adaptability and impact.