The Pictorial Trapezoid: Adapting McCloud’s Big Triangle for creative semiotic precision in generative text-to-image AI

Abstract

Generative artificial intelligence (AI) is rapidly being adopted in diverse research contexts that, given the specificity of theoretical frameworks and research objectives, require a high degree of semiotic precision in AI output. With text-to-image generative models, the selection of subject matter and subsequent stylistic variation both have the potential to influence measur- able desired outcomes. A major challenge in using generative models in design research is achieving a form of fidelity between a visual representa- tion and a corresponding concept that must be conveyed. Scott McCloud’s Big Triangle categorizes a broad range of visual representational stylistic variation, largely based on comic art. We extend the Big Triangle with a more systematically described framework called the Pictorial Trapezoid, which offers greater control in producing new pictures with generative AI. We provide a case study of the process by which we developed the Pictorial Trapezoid, and demonstrate its efficacy for an additional two research use cases. In each case we differentiate project-specific criteria for selecting what is being represented and visualizing that selection. Finally, we describe how an AI might be trained for semiotic precision in distinct research contexts using the Pictorial Trapezoid.