Image generation is a multimodal practice that extends far beyond the prompt or the final visual output. To understand this process, researchers must attend to the technological environments in which production takes place. This chapter unpacks the multimodal nature of AI image generation by distinguishing between the user interface and the underlying computational model. We argue that interfaces operate as semiotic surfaces whose affordances script what is actionable for the user. To analyse these dynamics, we introduce an adapted walkthrough method. As current approaches often struggle to capture the vast potential of generative tools, we propose a typology of three generative pathways (initiative, transformative, and referential generation). This framework allows researchers to systematically map possibilities beyond simple open-ended exploration, thus revealing how distinct modes of production are prioritised or constrained across different systems. Ultimately, our framework reveals that AI image production is a patterned practice: users do not simply generate images but work through structured pathways that reflect each platform’s particular design choices.