Research of the Influence of the Semantics and Syntax of a Text Request on the Image Quality Generated by Neural Network Techniques

The article deals with the problems associated with the generation of images by modern neural network techniques. The aim of the work is to develop and test methods that allow receiving correct generated images when entering a text request. The subject of the research is the modern neural network online services “Shedevrum” and “Kandinsky 2.2”. The relevance of the study is due to the introduction of neural network techniques in such areas of human life as PR, advertising services and even animation, theater and other arts. Both the main errors that occur when entering a text request in an online application or service, and ways to avoid them are given. To achieve this goal, the tasks of developing and testing an algorithm were solved that make it possible to formalize a text request, make it more “understandable” for interpretation by neural network services, which makes it possible to visually obtain more correct results of image generation compared to a natural language request. The methodological basis of the article is the method of system analysis, empirical, comparative and descriptive methods, as well as the method of generalization.