Studies / S.01

Less Siteless

2026

Traffic, solar, and pedestrian attraction in Suseong-gu, Daegu driven into the form itself through multi-objective optimization.

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 screen recordings with a Python capture script and processed through YOLO v.7, converted to tabular data for analysis. Solar exposure was quantified in Ladybug against the matching time bands, and pedestrian attraction was modelled as agents spawning by street and program category. Those readings became the input parameters for a Wallacei X optimization — image selector, rotation, and point picker against distance and displacement — which 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, and carried to plans, sections, and elevation.

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 →