Detecting the authenticity of news information published on social media has important practical significance. Due to the multimodal characteristics of false information such as information content, user related attributes, and information dissemination, existing false information detection technologies cannot effectively meet the multimodal detection requirements of false information. In this paper, we propose a different and novel method. Firstly, we propose a multimodal false information detection model that integrates multiple layers such as front-end, midrange, and back-end; Secondly, we fused the relevant field information of the existing public data set of false information with the generated content of GPT’s big model of artificial intelligence innovative learning ability in Natural language processing, expanding the public data set of false information; Finally, we utilized the extended false information public dataset based on GPT data fusion to train and validate this multi-layered fusion multimodal false information detection model. In actual detection, as long as the original tweet text content, user related attributes, and relevant evidence and other feature information are given, it can better predict whether the tweet news information is false information. Experiments have shown that compared with existing false information detection methods, the multimodal false information detection method based on knowledge distillation has significant accuracy advantages.