<p>Traditional knowledge has not fully utilized additional information such as entity images, relationship paths, and textual descriptions. Therefore, a cross domain multimodal knowledge graph adaptive embedding method based on text image enhancement is proposed. This method provides more comprehensive external information through mutual enhancement of text and images, compensating for the shortcomings of a single information source. Web crawling technology is used to collect cross-domain multimodal data for entity description and image modeling, thereby solving the problem of data sparsity and improving embedding effectiveness. We use multi-prototype text encoders and convolutional autoencoders to learn text and images online, model enhanced data, and obtain entity/relationship representations. Through joint training, an adaptive embedded knowledge graph is generated to enhance the capture of entity information and improve the representation quality and generalization ability of the knowledge graph. The experimental results show that this method effectively captures and enhances cross-domain multimodal data.</p>