A Generative AI Framework for Visual Pedagogy: Image-to Sketch Abstraction using Controlnet for Sketching Education

Abstract

Goal: This study addresses the challenges of teaching foundational sketching skills to novice engineering and design students by leveraging Generative Artificial Intelligence (GenAI). Rather than focusing solely on image creation, we propose a pedagogical framework that transforms complex reference images into simplified instructional outlines, aiding the internalization of visual structures. Design/Methodology: The pipeline integrates latent diffusion models with ControlNet spatial conditioning to generate customized visual outlines. This approach minimizes cognitive load for learners by extracting essential composition outlines while preserving the structural integrity of the reference image. Results: Expert evaluations show that the proposed GenAI method significantly outperforms traditional Canny edge-detection in compositional clarity, structural simplicity, and instructional utility. Contribution: This study introduces an innovative educational model, shifting the role of GenAI from an end product creator to a structured pedagogical assistant. It provides a practical roadmap for integrating GenAI into technical sketching curricula.