GAD Editorial · Published · · English study · Chinese summary · Study dated 2026-10-04
A roof option can look coherent from above while leaving its most important design work unresolved at the intersections. Where does one plane meet another? Which edge becomes a valley? What must happen below the silhouette? These questions make structured generation more interesting than the production of another attractive outline.
Roof-GAN, first submitted on 17 December 2020 and revised on 6 May 2021, generates residential roof geometry together with relationships between its primitives. The research represents both component shapes and their geometric connections. Its defined representation uses axis-aligned rectangular primitives with gable or hip configurations; wall height is held constant. These limitations belong to the research method and should stay visible when considering an architectural application. Paper and version history · Representation details
Read the relationship drawing
GAD proposes using the research as a prompt for an early roof-form workshop. Begin with a conventional project brief and a manually reviewed reference arrangement. Identify which decisions the exercise may explore, such as the composition of roof volumes, without assuming that generation will satisfy the actual footprint or project constraints.
For each candidate, draw a simple relationship diagram beside the perspective. Mark where elements meet, align or share a plane. Ask a second designer to reconstruct the intended arrangement from those records. If the relationship cannot be explained clearly, a polished rendering should not carry the option forward on its own.
Keep changes to the generator's output separate from the original candidate. Once a designer adapts a form to the real building, the result is a new project design with its own decisions. That adaptation effort matters when judging whether the research-inspired exercise was useful.
Follow the water, then the junction
Trace the proposed drainage direction across each plane and identify every collection point. This is an editorial design-review exercise, not a hydraulic analysis. Its purpose is to reveal where the team must resolve valleys, outlets, interfaces and maintenance access through the normal design process.
A visually simple intersection may become a demanding construction detail. Conversely, a composition with more visible parts may make the relevant interfaces easier to understand. Compare the unresolved junctions explicitly rather than ranking options by silhouette alone.
Include the spaces below. Ask how the roof affects the intended room volume, service zones and the coordination of openings. Keep these project requirements separate from what the research representation actually generates. A geometric relationship such as coplanarity does not establish a structural connection, weatherproofing detail or buildable sequence.
Evaluate what the exercise taught
Record which candidate introduced a useful possibility and which created additional unresolved work. The strongest outcome may be a revised design question rather than a selected generated roof.
This is an application framework derived from historical research, not a hands-on Roof-GAN review. Before assigning it any production role, a practice would need to test the implementation and its suitability for the building type. For early design discussion, its central contribution is already legible: a roof is a system of relationships, and those relationships deserve to be reviewed before the image becomes persuasive.