Researchers at Cornell Tech and the Cornell Bowers College of Computing and Information Science have introduced ShadowDraw, an artificial intelligence framework designed to generate composite shadow art from digital scans of everyday objects. The system captures a physical item, projects its cast shadow, and algorithmically generates a line drawing that merges seamlessly with the shadowed region. This automated workflow eliminates the manual trial and error traditionally required for shadow art, delivering a complete artistic composition in a single computational step. Lead author and doctoral candidate Rundong Luo developed the framework alongside advisors Wei-Chiu Ma and Noah Snavely. The team drew inspiration from the handcrafted shadowology illustrations of Belgian artist Vincent Bal. Recognizing that manually produced shadow art lacks sufficient volume for machine learning training, Luo pivoted to a scalable dataset strategy. The model was trained on extensive online collections of line drawings, learning to predict full sketches conditioned on the closed geometric boundaries formed by an object’s shadow. Testing demonstrated the framework’s versatility across single and multiple object scans. The system also supports animation by rendering sequential key frames and overlaying evolving shadow contours onto a single composition, utilizing distinct color channels to differentiate temporal stages. Ma noted that the project successfully bridges two-dimensional AI processing with three-dimensional physical objects, extending computational creativity beyond digital interfaces. The researchers explicitly position ShadowDraw as a collaborative instrument intended to augment human creativity rather than replace professional artists. Luo will present the methodology and findings in the paper ShadowDraw: From Any Object to Shadow-Drawing Compositional Art at the Computer Vision Foundations Conference on Computer Vision and Pattern Recognition. The conference will convene in Denver from June 3 to June 7. The framework establishes a new pipeline for transforming physical objects into structured visual narratives through automated generative design.