Designing with Latent Space: A Semantic Canvas and Agent to Broaden Exploration and Clarify Intent in Early Stage Footwear Design

This thesis investigates whether latent space can be reframed as a designer-facing medium through a Semantic Canvas, which projects generated images onto designer-defined semantic axes, and a Brief-Aware Agent, which maintains the design brief and surfaces vocabulary, alternatives, and axis suggestions. The work responds to a familiar working condition in generative design tools: visual design thinking has to pass through natural language before reaching the model, and designers commonly fixate on early high-fidelity outputs and lose track of how intent evolves. Deployed as a situated technology probe at HILOS Studio with five professional footwear designers across structured sessions and self-directed use, the study finds that making latent structure visible makes fixation legible rather than removing it, and that distributing design intent across multiple channels carries more of what the designer thinks than text prompts alone. The system is offered as one direction for future AI-augmented creative tools, suggesting that the interface layer above a generative model can take on some of the work of holding intent and surfacing fixation.