Mobile AIGC Image Generation Interactive Training Model for Design Education

Abstract

In the field of design education, the current AI-generated content (AIGC) tools commonly face key issues such as poor adaptation to mobile platforms, disconnection between interaction logic and educational objectives, and the inability to support creative iteration with generated results. To address these challenges, this paper proposes a mobile AIGC image generation interactive training model tailored for design education scenarios. The research first constructs a three-layer architecture that includes multimodal interactive input, lightweight model generation, and educational feedback iteration, based on the core needs of creative inspiration, solution iteration, and professional feedback in design education. Then, through techniques such as knowledge distillation, quantization compression, and operator optimization, the Stable Diffusion model is improved and adapted for mobile platforms, achieving a balance between generation quality and resource consumption. Finally, an interactive mechanism is designed based on constructivist educational theory, supporting professional design inputs such as sketches, text, and color references, and incorporating design guidelines and creativity-expanding modules. To evaluate the model’s performance, mobile devices of various specifications (low, medium, and high-end) are tested in three typical educational scenarios: 1) graphic design, 2) product design, and 3) user interface (UI) or user experience (UX) design. Comparison experiments are conducted with existing mobile AIGC models, such as mobile stable diffusion and lite generative adversarial network (GAN), using performance indicators such as Fréchet inception distance (FID), inception score (IS), educational adaptability score, and resource consumption metrics. The results show that the proposed model outperforms existing models in core performance metrics, with an average generation speed improvement of 42.3%, an educational objective match score of 8.7/10, and a 31.6% reduction in mobile memory usage. The findings provide effective technical support for creative generation, solution iteration, and personalized teaching in design education.