Abstract Idioms have various compositions and are part of the long history and culture of the Chinese nation, with their unique ways of expression. In this paper, taking “Pictures of Idioms” as an example, we propose a strategy for designing illustrations for a related series of books on idioms using artificial intelligence technology. Using the LSTM model, the description of the given idiom text is realized to generate images. The generated images are processed by a decoder and fed into a neural network model to generate illustrations that match the description of the idiom text. Further, a diffusion model is proposed to improve the quality of illustration generation and diversify the illustrations by continuously adding conditions to the generation process. Taking “Picture Talking Idioms” as an example, we propose a design strategy for the illustrations of the Idioms series accompanying elementary school language textbooks and quantify the effect of practical application in terms of the design effect of the final product, the reading satisfaction, and the effect of the use of “Picture Talking Idioms.” Students were more satisfied with the use of blue-gray and blue colors in the illustrations in “Diagrams to Tell Idioms,” with ratings of 4.9 and 4.6, respectively. 52% of the students used the diagrammatic method of reading “Diagrams to Tell Idioms” and rated the knowledge content of the books in terms of idioms at 9.458, which is a high satisfaction rate of 98%. The use of diagramming idioms has been very effective in assisting with language text.