Victor Sardenberg

In May 2026, at the Digital Methods of Engineering (DME) chair of Bergische Universität Wuppertal, we explored how deeply AI can join the creative process when its output returns not as an image but as geometry. Instead of starting from parametric models, participants anchored their designs in cultural references — albums, paintings, sculptures, films — which became text prompts for Stable Diffusion; the generated images were then converted into 3D meshes with Hunyuan3D 2.1 inside ComfyUI, imported into Rhino, edited, and fed back as new Canny and depth maps to condition the next round of generation. Slower and messier than a purely parametric loop, this image-to-3D feedback shifted AI from a source of visual inspiration to a co-author of spatial form, turning imperfect meshes — ambiguous in scale, rough in topology — into provocations that students negotiated and reinterpreted rather than corrected.