Integrating Flipped Learning and Generative AI for Practice-Based Design Education: Evidence from a Knit Yarn Design Course

1 Introduction

• Material, equipment, time, and support constraints can make iteration costly, narrowing exploration and weakening the feedback–revision loop.This challenge is especially relevant where learning requires repeated physical trials.Source: §1 · p. 2
• Uneven procedural knowledge and understanding of assessment expectations can consume studio time and delay experimentation and critique.The concern is particularly acute in material-oriented design courses.Source: §1 · p. 2
• Evidence on GenAI in artifact-centered design courses remains limited, where outcomes must account for creative intention and practical feasibility.Knit yarn design requires students to coordinate yarn properties, stitch structures, techniques, and repeated sampling.Source: §1 · p. 3
• The study proposes an undergraduate knit yarn design framework connecting flipped preparation, exemplar review, GenAI-supported ideation, and studio refinement.The cross-device platform included pre-class videos, embedded checks, a curated gallery, and optional parameterized prompt scaffolds; GenAI supported ideation rather than final assessed artifacts.Source: §1 · p. 3
• The evaluation compared a prior cohort without the intervention with an intervention cohort using the platform, while treating historical-cohort findings as exploratory rather than causal.The same instructor taught both cohorts, and assessed components and weighting were held constant.Source: §1 · p. 4
• The study’s pedagogical contribution is a framework that assigns complementary roles to preparation, visual exploration, hands-on making, and instructor judgement.The authors suggest it may inform design courses combining visual ideation, material experimentation, and technical development.Source: §1 · p. 4

2 Platform Implementation

• The responsive platform connects pre-class learning, studio readiness, and early visual ideation across desktop and mobile devices.It was designed to support continuity between preparation, classroom activity, feedback, and later sampling.Source: §2 · pp. 4–5, Figure 1 · p. 4

Course module

• Eight approximately five-minute micro-videos and formative readiness checks supported foundational preparation alongside continuing teacher-led lectures.Checks followed video segments, allowed retries, and provided correctness feedback with explanations; the fourth lecture included student presentations linking preparation to later studio work.Source: Course module · pp. 5–6, Figure 2 · p. 5

Gallery module

• The Gallery organized feasible yarn-design exemplars across thirteen lab-producible yarn categories to calibrate quality expectations.It included selected prior-cohort work but excluded artifacts from the historical control cohort used in the comparison.Source: Gallery module · pp. 6–7, Figure 3 · p. 6

GenAI module

• The GenAI module scaffolded prompts using course-relevant parameters and optional reference images to support early ideation, not technically accurate or manufacturable outcomes.Students could vary color, yarn type, stitch structure, texture, material effect, and style; generated images required interpretation and later hands-on testing. The module also offered a 10-color palette with hexadecimal codes from uploaded images.Source: GenAI module · pp. 7–8, Figure 5 · p. 8

Instrumentation and learning traces

• Platform logs captured video engagement, readiness checks, gallery visits, and GenAI interactions as behavioral indicators of preparation and exploration.GenAI records included generation frequency, parameter selections, and related interaction data.Source: Instrumentation and learning traces · p. 9

3 Methodology

• Figure 6 summarizes the study design and data-collection procedure.The supplied passage identifies the figure’s scope but does not specify its individual stages.Source: Figure 6 · p. 9

3.1 Study design

• The historical-cohort comparison included 12 students in the prior-year control and 16 in the current-year intervention cohort.Both were third-year students taught by the same instructor, following the same brief, deliverables, rubric, and grading scheme; the intervention cohort used the platform-supported flipped-learning arrangement and GenAI ideation module.Source: §3.1 · pp. 9–10
• Because cohorts were not randomly assigned and were drawn from different academic years, cohort differences and contextual factors cannot be ruled out.The evaluation is exploratory and reports descriptive statistics, effect sizes, confidence intervals, and non-parametric sensitivity tests where appropriate.Source: §3.1 · p. 10

3.2 Course context, rubric, and learning outcomes

• Five rubric dimensions—Creativity thinking, Design skills, Problem solving, Presentation, and Attendance—were scored on a 0–100 scale.Their weightings determined the total course score.Source: §3.2 · p. 10
• The same instructor scored both cohorts using the same rubric and weighting scheme.Source: §3.2 · p. 10

Pre-class learning arrangement

• The intervention platform provided pre-class videos, learning materials, examples, and formative checks about yarn production, machine operation, yarn selection, and knitting procedures.These resources were intended to prepare students before entering the studio.Source: Pre-class learning arrangement · p. 10
• Pre-class preparation was intended to free studio time for demonstration, hands-on practice, discussion, experimentation, and feedback.Source: Pre-class learning arrangement · pp. 10–11

GenAI-assisted ideation module

• The GenAI module supported early-stage exploration, comparison of visual directions, and reflection on possible yarn-design variations.Source: GenAI-assisted ideation module · p. 11
• The GenAI module was a support tool for generating and comparing visual references, not a replacement for instructor feedback, hands-on demonstration, or technical decision-making.Source: GenAI-assisted ideation module · p. 11

Course scores

• For both cohorts, instructor-assigned rubric-dimension scores and weighted total course scores were recorded on a 0–100 scale.Source: Course scores · p. 11

Questionnaires

• Intervention students completed pre-course and post-course questionnaires using 5-point Likert-type items and optional open-ended questions.The pre-course questionnaire captured perceived baseline dispositions and abilities; the post-course questionnaire evaluated the learning experience, including flipped learning and GenAI.Source: Questionnaires · pp. 11–12
• Pre–post comparisons covered Innovation and Motivation and Self-Directed Learning Ability, but differently worded items measured shifts in self-reported perceptions rather than strict latent-construct growth.Source: Questionnaires · p. 11

Brief interviews

• Brief interviews with four intervention students and the instructor explored preparation, studio learning, GenAI-supported ideation, and the overall course setting.Interview summaries were used as illustrative context for quantitative patterns.Source: Brief interviews · p. 12

3.5 Data analysis

• Course scores and questionnaire responses were summarized with means and standard deviations, with favorable questionnaire responses defined as ratings of 4 or 5.Source: §3.5 · p. 12

Between-cohort comparison of course performance

• Welch’s two-sample t-tests compared the historical control and intervention cohorts on the weighted total and five rubric dimensions, with differences calculated as intervention minus control.Hedges’ g, bootstrap 95% confidence intervals from 5,000 resamples, and Mann–Whitney U sensitivity analyses were also reported.Source: Between-cohort comparison of course… · p. 12

Pre–post questionnaire comparison

• For intervention students, construct scores averaged four items, and pre–post changes were analyzed as post-course minus pre-course scores using Wilcoxon signed-rank tests.Because items were aligned but not identically worded, changes were treated as exploratory indicators; bootstrap confidence intervals and favorable response rates were also reported.Source: Pre–post questionnaire comparison · p. 12

Interview and open-ended response analysis

• Post-course ratings of flipped learning, the GenAI module, and the overall setting were analyzed descriptively, while interviews and open responses were reviewed for recurring learning-experience themes.Given the small qualitative dataset, interpretation was descriptive and contextual rather than an attempt to generate generalizable findings.Source: Pre–post questionnaire comparison · p. 13, Interview and open-ended response… · p. 13

4.1 Between-cohort comparison of course performance

• 4.87 points (95% CI [2.16, 7.48]) was the total-score mean difference; creativity thinking, design skills, and problem solving were significant, while presentation and attendance were not.Mann–Whitney sensitivity tests showed a broadly consistent pattern, with significant differences in creativity thinking, design skills, problem solving, and total course score.Source: Table 2 · p. 13, §4.1 · pp. 13–14, Figure 7 · p. 14

4.2 Pre–post changes in questionnaire constructs

• Mean change was 0.52 for Innovation and Motivation and 0.48 for Self-Directed Learning Ability; non-identical item framing makes these perception comparisons, not direct evidence of latent growth.Both pre–post differences were statistically significant in the intervention cohort.Source: §4.2 · p. 15, Table 3 · p. 15

4.3 Post-course evaluation of the learning setting

• Section-level ratings were 4.17 for flipped learning, 4.00 for the overall class evaluation, and 3.63 for the GenAI module.Students rated the overall learning setting positively while rating its components somewhat differently.Source: §4.3 · p. 15, Figure 8 · p. 16
• Flipped-learning items ranged from 4.00 to 4.25, while GenAI ratings were 4.06 for idea expansion but 3.00 for time-saving and 2.94 for direct use of AI references.The GenAI ratings were stronger for ideation-oriented items than for time-saving or direct-output-use items.Source: §4.3 · pp. 15–16

4.4 Qualitative findings from student and instructor interviews

• Students described pre-class videos as helping them understand machine operation, yarn processes, selection, and procedures before studio work.Interviews also linked preparation with clearer procedural knowledge and broader understanding of yarn-making principles.Source: §4.4 · pp. 17–18
• The instructor observed earlier completion of basic samples and more student effort in documenting, presenting, and reflecting on design processes.Some students finished two samples by Week 8 or 9 rather than around Week 11, leaving more time for experimentation and refinement.Source: §4.4 · p. 18
• Students used GenAI mainly to explore visual directions; final designs continued to rely on instructor feedback, material testing, and hands-on studio work.Source: §4.4 · p. 18
• A gradient swatch case showed iterative problem-solving: the student recalculated production parameters after the initial sample missed the intended effect.The revised sample better approximated the intended gradient, illustrating iterative reasoning rather than generalizable evidence.Source: Figure 9 · p. 17, §4.4 · pp. 18–19

• The clearest outcome differences were in creativity thinking, design skills, problem solving, and total course score; presentation and attendance differences were weaker or non-significant.Source: §5 · p. 19
• The framework assigns complementary roles to preparation, feasible examples, early visual exploration, hands-on making, and studio feedback.Feedback and studio practice support feasibility judgment, parameter adjustment, and material realization.Source: §5 · p. 19
• Flipped preparation may reduce uncertainty before studio work, while GenAI mainly scaffolds early visual exploration rather than providing technical advice.Students explored color, yarn, structure, texture, and visual effect; technical judgment remained grounded in instructor feedback and studio practice.Source: §5 · p. 19
• A gradient-knit swatch anecdote illustrates iteration from a mismatch between intended and physical outcomes through instructor consultation and parameter revision.The authors characterize the case as anecdotal and describe the resulting iteration as consistent with the reasoning the course aimed to support.Source: §5 · p. 19
• Historical cohorts without random assignment leave cohort or contextual factors as possible influences; the small cohort limits statistical power and generalizability.The limitation is especially relevant to questionnaire and interview findings.Source: §5 · pp. 19–20

6 Conclusions

• Across cohorts, the intervention was associated with higher performance in creativity thinking, design skills, problem solving, and total course score.Source: §6 · p. 20
• Students in the intervention cohort reported positive pre-post shifts in Innovation and Motivation and Self-Directed Learning Ability.Source: §6 · p. 20

Declaration of generative AI and AI-assisted technologies in the writing process.

• The authors used ChatGPT/Codex for language editing, structural revision, and clarity checking, then reviewed and edited the content and accepted responsibility for it.Source: Declaration of generative AI and… · p. 21

• Tables A1 and A2 are identified as the pre-survey and post-survey questionnaire item tables.Source: Table A1 · p. 26, Table A2 · p. 27
• The survey instructions specify a five-point Likert scale from Strongly Disagree to Strongly Agree.The pre- and post-survey instructions phrase the response task differently.Source: Appendix A · pp. 26–27
• Post-survey items cover overall class-setting evaluation, innovation and motivation, self-directed learning, and suggestions for improving the AI agent.Source: Appendix A · p. 28
• Figures A1, A2, and A3 present item-level Likert distributions for perceived flipped-learning effects, AI-agent effects, and the overall learning setting, respectively.Source: Figure A1 · p. 29, Figure A2 · p. 30, Figure A3 · p. 31