One of the trends in the current stage of digital transformation of the education system is the intensive use of large generative models. Various aspects of the problem of applying artificial intelligence systems in pedagogy are the subject of numerous studies, however, despite this, the practical aspects of using neural networks for image generation are still relatively underexplored. The aim of the study is to analyze the role, place and potential of using neural networks for image generation in teaching students to work with digital design. The authors used theoretical and empirical research methods. A comprehensive review of scientific works by domestic and foreign authors allowed us to conclude that the primary attention is paid by researchers to the personalization and individualization potential inherent in neural networks and the associated ethical issues. The article presents the experience gained in using neural networks for image generation within the framework of the course "Modern Visualization Technologies in Marketing". It considers the content of the course and methods of its study. The results of the survey among students who have studied the course were analyzed. The survey aimed to gather information about the existing experience with working with neural networks for image generation, the convenience of using specific neural networks, the quality of generated images, and any problems that may have arisen while working with neural networks. The article discusses the use of chatbot-based writing prompts by students, and the technologies involved in this process. An analysis of the survey results revealed that the key skill required for the successful use of neural networks is the creation of effective prompts. The Recraft neural network was rated as more suitable for teaching digital design than the Kandinsky network. The theoretical significance of the study lies in its analysis of the specific features of using large generative models in the educational process. A scenario for the multimodal use of large generative models by students was identified, in which one neural network acts as a prompt generation tool for another neural network. The practical significance of the study lies in the development and practical testing of the content, forms, and methods of using neural networks for image generation in digital design.