Generative AI configurations across creative phases: a 2 × 2 experiment on creative performance and immediate knowledge in art and design education

Introduction Generative artificial intelligence (GenAI) is increasingly used in art and design education, but it remains unclear whether text-based and image-generative configurations offer different advantages across creative phases or whether sequencing them improves final outcomes. Methods In a randomized 2 × 2 between-subjects experiment, 120 art and design students (30 per condition) completed a standardized cultural-heritage chromatic reconstruction task. Participants were assigned to Text-to-Text, Text-to-Image, Image-to-Text, or Image-to-Image configurations across divergent exploration and convergent development. Expert-rated creative performance, immediate task-relevant knowledge, and human–AI interaction processes were assessed. The four primary hypothesis tests were evaluated with Holm correction. Results Text-based AI during divergence produced higher divergent creative performance than image-generative AI (mean difference = 0.823, Holm-adjusted p < 0.001). Image-generative AI during convergence yielded higher Aesthetic Quality at the unadjusted level (difference = 0.493, p = 0.026), but not after Holm correction (adjusted p = 0.052). The hypothesized Text-to-Image sequence did not significantly outperform alternatives in Overall Creative Performance (contrast = 0.066, p = 0.671), and the factorial interaction was nonsignificant ( p = 0.541). However, Text-to-Image achieved higher baseline-adjusted immediate post-task knowledge scores (contrast = 1.478, 95% CI [1.050, 1.906], p < 0.001), without a significant cross-phase interaction ( p = 0.404). Discussion GenAI configuration effects differed across creative phases and outcomes, but phase-specific advantages did not establish a superior final creative-product sequence. Because AI modality was bundled with interaction and interface characteristics, and the knowledge assessments had limited reliability, the results do not establish isolated modality effects, durable learning, or synergistic sequencing. AI orchestration should remain a provisional, context-sensitive instructional perspective requiring further validation.