Semantic Prompt Interface (SPI): A Meaning-Oriented Interaction Framework for Generative Image Systems

Prompt-based image generation has become the dominant interaction paradigm in generative AI systems, yet it remains heavily constrained by text-centered command structures. This paradigm positions users as instruction-givers and AI systems as executors, making it difficult for users to understand and control how semantic intent is interpreted and translated into visual outcomes. This work introduces the Semantic Prompt Interface (SPI), an interaction framework that reconceptualizes prompts not as fixed textual commands, but as collections of manipulable semantic units and relationships. SPI exposes intermediate semantic structures derived from an initial prompt and enables users to explore and adjust how meaning is interpreted during the generation process. By reframing prompt-based image generation as an interaction design problem rather than a prompt-writing skill problem, SPI shifts human–AI interaction toward co-interpretation and meaning-oriented exploration. This paper positions SPI as a conceptual and interface-level contribution, highlighting the role of generative AI systems not only as production tools, but as interactive environments that support and refine human thinking.