ARCHITECTURELM
Index
Tools
Spaces
About
No. 001
August 27, 2026
We ran the same Rhino massing through five settings to find where geometry survives and where it quietly dissolves.
By Ahmed Salah
Fig. 01 — Flux.2 Klein · v2.1 · presentation stage · 40s · A. Salah
Every rendering tool promises to respect your geometry. Few do. We tested a single mid-rise massing model exported from Rhino through Flux.2 Klein at five denoising strengths, holding prompt and seed constant, to establish where the model stops interpreting and starts inventing.
A nine-storey residential block with a deep colonnade at ground level and a regular fin rhythm above. The colonnade matters: it is the kind of feature that AI rendering tools routinely flatten into a painted-on shadow.
Below 0.45 strength, the fin rhythm survived intact. Bay spacing stayed within a fin of the source model, and the colonnade read as real depth rather than a graphic.
Above 0.6, the model began adding storeys. Not dramatically, but a nine-storey block returned as ten in three of eight generations. For competition imagery this is disqualifying.
For presentation work where the geometry is fixed, 0.35 to 0.5 is the usable band. Higher settings belong to early concept work where the massing is still a suggestion.
PUBLICATION
ArchitectureLM
ISSUE
014 — Rev C
EDITION
2026
CONTACT
hello@architecturelm.com