Less Siteless — Brad Machado

Type: Graduate design studio — midterm
Role: Group project
Tools: Grasshopper · Wallacei X · Ladybug · YOLO v.7 · Flux.1-dev LoRA · ComfyUI · Python · 3D printing
Status: Completed — carried into Artificial Context

Cluster of white 3D-printed unit forms on steel pins, mounted to an ash plywood base plate, photographed at an angle. — Fig. 01 — Printed unit field, ash base. 1:50.

01 / Process

Method

The intent is to create form that is “less siteless” — i.e. contains context within the form. Site imagery of Suseong-gu, Daegu was gathered and used to train a Low-Rank Adaptation model on Flux.1-dev in ComfyUI; the generated images were then rated against three categories — technology, education, and quality of life — which returned an overwhelming 91.97% weighting toward quality of life.

Three grayscale renders of twisted, spiraling surface forms shown side by side as sequential form studies. — Fig. 02 — Form studies, twisted surface variations.

Wireframe render of an undulating surface intersected by a flat cutting plane, with thin contour lines marking the intersection. — Fig. 03 — Surface and cutting-plane study, contour extraction.

02 / Outcome

Result

Vehicle movement was extracted from CCTV recordings through YOLO v.7, solar exposure was quantified in Ladybug, and pedestrian attraction was modelled as agents spawning by street and program category. Those readings drove a Wallacei X optimization that selected the lofted form with the least displacement at each floor. The form was resolved into concrete structure, floor plates and a circulation core, with a facade generated by pouring liquid over the lofted geometry.

Grayscale Grasshopper visual-scripting canvas showing the definition used to generate and cut the surface forms. — Fig. 04 — Grasshopper definition, form generation and cutting logic.

Next Project Artificial Context →