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

Qu, Ling, and Yang (2026) report an instructional framework that joins flipped learning, an exemplar gallery, and GenAI-assisted visual prototyping in an undergraduate knit yarn design course. In practice-based studio work, visual intentions must become feasible material outcomes, and in knit design those decisions are bound to yarn properties, stitch structures, machine operation, and the cost of physical sampling.

The authors built a cross-device platform with pre-class micro-videos, embedded readiness checks, a curated gallery, and a GenAI ideation module, then evaluated it with a quasi-experimental historical-cohort design: a historical control cohort (n = 12) from the previous year's offering against an intervention cohort (n = 16), same course, same instructor, same rubric weighting — Creativity thinking 35%, Design skills 20%, Problem solving 20%, Presentation 15%, Attendance 10%, scored on a 0–100 scale. Questionnaire responses and brief interviews with four students and the course instructor supplemented the score comparison, and the authors read the quantitative results as exploratory indicators rather than causal evidence.

Synthesis: This study documents GenAI inside an artifact-centered studio course, where learning is judged by what students can actually make. Its contribution is pedagogical: flipped preparation and readiness checks lower uncertainty before machine time, the gallery calibrates what counts as feasible, GenAI widens the early visual search space, and instructor critique plus physical sampling retain authority over technical judgment. Read with its small, non-randomized cohorts in view, the pattern of higher creativity thinking, design skills, problem solving, and total course score is best treated as a hypothesis for other Design Education courses that must reconcile visual ideation with material feasibility.

Key Findings

1. The weighted total course score was 80.04 (SD = 4.12) in the historical control cohort and 84.91 (SD = 3.29) in the intervention cohort, a difference of 4.87 points [2.16, 7.48], p = 0.003, Hedges' g = 1.29.
2. Creativity thinking moved from 79.50 (SD = 3.97) to 84.38 (SD = 4.59), a difference of 4.88 [1.81, 8.02], p = 0.006, g = 1.09.
3. Design skills moved from 78.42 (SD = 3.40) to 84.13 (SD = 4.24), a difference of 5.71 [2.98, 8.35], p < 0.001, g = 1.42.
4. Problem solving moved from 78.58 (SD = 5.85) to 84.94 (SD = 4.30), a difference of 6.36 [2.40, 10.00], p = 0.005, g = 1.23, while presentation (p = 0.100, g = 0.62) and attendance (p = 0.445, g = 0.29) did not separate the cohorts.
5. Innovation and Motivation rose from 3.64 (SD = 0.55) to 4.16 (SD = 0.70), a change of 0.52 [0.23, 0.83] with favorable responses up from 60.9% to 82.8% (p = 0.0072), and Self-Directed Learning Ability from 3.59 (SD = 0.61) to 4.08 (SD = 0.72), a change of 0.48 [0.22, 0.75] with favorable responses up from 59.4% to 78.1% (p = 0.0058).
6. Section ratings favored the flipped-learning component (mean = 4.17) and the overall class setting (mean = 4.00) over the GenAI module (mean = 3.63); GenAI scored highest for expanding creative boundaries and offering ideas students might not otherwise have considered (both 4.06) and lowest for saving time in experimentation and preparation (3.00) and direct incorporation into final outcomes (2.94).
7. The instructor reported that where students previously needed until around Week 11 to complete two samples, many finished two samples by around Week 8 or Week 9, and some were then more willing to try a third sample.

Why GenAI helps early and stops there

The GenAI module called a commercial image generation model through an API and let students assemble prompts from the course's own design parameters: color direction, yarn type, knit or stitch structure, texture, material effect, and visual style, with one to three optional reference images and a palette tool that produced a 10-color palette with hexadecimal codes. The authors are explicit that these images were ideation prompts, not simulations of textile outcomes — current models have limited understanding of yarn behavior, stitch formation, knit structure, material properties, and workshop feasibility, and could not represent executable knitting instructions. The ratings match that framing: strongest for broadening the early search space, weakest for time saving and for feeding assessed artifacts.

Preparation as the real lever

The flipped arrangement moved yarn production, knitting processes, machine operation, and yarn selection out of studio time through two lessons containing eight micro-videos of roughly five minutes, each pair of segments gated by a viewing threshold before its readiness checks unlocked and with bypass functions disabled. Students described entering the workshop with the sequence already in mind, which let them concentrate on difficult steps instead of following instructions passively, and connected spinning principles to how they might design later. In the fourth lecture session students explained what they had learned to the class, an activity the authors credit with reinforcing technical understanding.

What the score pattern does and does not say

Three of five rubric dimensions plus the total score separated the cohorts with large standardized differences, while presentation and attendance did not — which the authors read as the intervention acting on the design-generative parts of the course rather than on generic participation. Their gradient knit swatch case shows the mechanism: a student whose sample missed the intended gradient, because the effect depended on swatch size and related production parameters, consulted the instructor, re-examined the parameters, recalculated the design plan, and produced a revised sample closer to the intended effect. The instructor also saw more evidence of reflection in portfolio layouts and process documentation than in previous years.

What this means for practice

• Instructors. Keep GenAI in the ideation stage and say so in the brief: have students bring generated references to critique as prompts for comparison rather than as targets, while instructor feedback and physical sampling carry judgment about yarn properties, stitch structure, and machine constraints. Gate the pre-class checks behind a viewing threshold and give correctness feedback with brief explanations.
• Instructional designers. Build the gallery deliberately. The authors drew exemplars from earlier teaching materials outside the two compared cohorts, excluding the control cohort's assessed artifacts, which protects the comparison while still calibrating quality expectations.
• Tool builders. Parameter-guided prompt scaffolding that speaks the domain's vocabulary lowers the prompt-writing barrier, but generated images should not be marketed as a feasibility check for stitch construction, tension, or scale.

Limitations

• The comparison is a quasi-experimental historical-cohort design with no random assignment: 12 students from the previous year's offering against 16 in the intervention year, so cohort differences and other contextual factors cannot be ruled out.
• It rests on one course at one institution, and all course scores were assigned by a single instructor using a common rubric, so scoring is not independent of the person who designed and delivered the intervention.
• The questionnaire was collected from the intervention cohort only, and its pre- and post-course items were conceptually aligned but not identical repeated measures, so the shifts are self-reported perceptions rather than psychometric evidence of latent growth.
• The qualitative material is thin by design: brief interviews with four students and the instructor, analyzed interpretively, with the swatch case presented as anecdotal.

Connected Concepts

• Design Education
• Generative AI
• Experiential Learning
• Scaffolding
• Creativity
• Self-Directed Learning

Connected Articles

• Development and applications of Generative AI in architectural design studios — Generative AI in architectural design studios
• Generative AI in Design Thinking Pedagogy: Enhancing Creativity, Critical Thinking, and Ethical Reasoning in Higher Education — GenAI and design thinking in design education
• Cultivating Design Creativity of Vocational Students: A Model of Project-Based Learning in AI-Enabled Immersive Virtual Environments — GenAI-supported project-based learning and creativity in vocational design

Citation

Qu, H., Ling, Z., & Yang, Y. (2026). Integrating Flipped Learning and Generative AI for Practice-Based Design Education: Evidence from a Knit Yarn Design Course. arXiv preprint arXiv:2609.18505.