Work

A Guided Prompting Interface for Supporting Creative Intent in AI Image Generation

Year: 2026
Type: article
Abstract: Background : Recent advances in generative artificial intelligence (AI) have expanded the accessibility of text-to-image creation. However, novice users still face substantial challenges when using such systems. Beginners often struggle to translate abstract visual intentions into effective prompts due to unstructured prompt composition, difficulties in articulating abstract visual attributes, and the unpredictability of generated outcomes. Accordingly, this study aims to propose a guided prompt interface designed to systematically support novice users in structuring and controlling their creative intent during AI-based image generation. Methods : This study employed a combination of formative and experimental research approaches. First, a formative study involving 16 novice users was conducted using observational tasks and in-depth interviews to identify key difficulties and user needs in existing text-to-image generation tools. Based on these findings, prior literature, and an empirical analysis of 100,000 real-world user prompts, a six-category framework—Subject, Style, Artistic Reference, Composition, Mood, and Tone—was developed to define the core components of effective prompts. An interface integrating a step-by-step guide, visually represented filters, and a generative pre-trained transformer (GPT)-assisted prompt recommendation feature was then designed based on this framework. Finally, a comparative user study with 16 participants was conducted to evaluate the proposed interface against Midjourney using Computer System Usability Questionnaire (CSUQ) and qualitative interaction analysis. Results : The formative study revealed that novice users experienced difficulties in structuring prompts, expressing abstract visual attributes, and anticipating system outputs. The proposed method addressed these challenges by enabling users to construct prompts using 65 visually represented filters organized within a step-by-step workflow. Comparative evaluation results showed that the proposed interface significantly outperformed Midjourney in overall usability, particularly regarding ease of use and clarity of information organization. Qualitative findings further indicated that the structured interface improved users’ confidence, reduced trial and error, and enhanced the predictability of generated images by aligning them more closely with users’ intended visual characteristics. Conclusions : This study builds upon existing AI image generation technologies by proposing a unified, step-by-step interface that structures creative intent and enhances user control throughout the image generation process. By reducing cognitive burden, the proposed interface helps novice users articulate and manage their creative decisions more clearly and effectively. Future research should incorporate open-ended creative tasks and longitudinal user evaluations to further investigate how structured guidance can coexist with creative flexibility in human–AI collaboration.
Language: English
Open Access status: diamond
Award: G431652139
OpenAlex ID: W7131237349
FWCI: 5.425
Cites: 36
Cited by: 2
Related to: 10