I get a more reliable city model when I separate exploration from assembly. Shapezo is useful for generating a spatial hypothesis from a map-selected area. CADMapper is useful for converting AutoCAD and related layers into a structured 3D city context. The pipeline works only when I keep the status of each asset obvious.
Shapezo produces inferred geometry. CADMapper works from source-backed drawings and data. Neither label is a criticism. They describe what a model can safely be used for, and they determine the checks I need before export or analysis.
1. Define a bounded experiment
I start with one decision and one boundary: a station block, industrial parcel, campus edge, or waterfront segment. In Shapezo, I test massing, access, density, and open-space ideas inside that boundary. I save the boundary, prompt assumptions, and date with every concept so the experiment can be repeated.
At this point I do not expect reliable object semantics. An AI-generated road may be plausible without matching a right of way. A building may look reasonable without respecting a survey or utility corridor. I mark the output as concept-only and use it to decide which questions deserve source data.
2. Prepare the CAD and GIS inputs
When an option is worth developing, I prepare CADMapper inputs. Depending on the project, that can include DWG, DXF, PDF, SVG, terrain, imagery, and GIS layers. Before conversion I check coordinate reference, units, layer meaning, coverage, source dates, and the level of detail required by the next tool.
This preparation is not busywork. A city model inherits the assumptions in its source layers. If a road is offset, a parcel is in the wrong coordinate system, or a terrain surface is missing, the resulting 3D scene can be precise and still be wrong.
3. Convert and refine by decision
CADMapper helps turn footprints, alignments, buildings, roads, terrain, utilities, and public-space elements into a connected city model. I refine in priority order instead of trying to detail everything. For an access study, roads, grades, entrances, and service areas come first. For a view study, building heights, roofs, and surrounding blocks matter more.
The same scene can support different exports, including CityGML, OBJ, PLY, STL, Collada, or 3D PDF. I select the format after checking the consumer. A semantic city exchange needs different information from a lightweight WebGL visualization or a render asset for Lumion.
4. Validate before analysis
A converted model is not automatically an analysis model. Before I use it for daylight, viewshed, wind, movement, land-use, or energy questions, I check scale, coordinate alignment, terrain quality, missing buildings, object completeness, and source dates. I also confirm that the level of detail is appropriate; a public-facing scene and a simulation input rarely need the same geometry.
For repeatable builds, I keep conversion settings beside the source package. Layer filters, height rules, simplification choices, and export options can change the scene just as much as a new drawing. Recording them means another developer can reproduce the result and identify whether a revision came from new data or from a changed rule.
Failure modes I watch for
The first failure mode is confusing a map boundary with a complete site description. The second is treating AI massing as survey or engineering truth. The third is exporting without checking coordinate systems and destination requirements. I avoid these by keeping three explicit states: generated concept, source-backed city context, and validated analysis input.
I also keep a small layer manifest. It records which source produced each major object, which rule modified it, and what level of detail it carries. When a reviewer asks why a building, road, or water surface differs from the drawing, I can trace the change instead of guessing.
My implementation rule
Use Shapezo to decide what deserves modeling effort. Use CADMapper to build the connected context that lets the decision be reviewed. Keep the handoff traceable, keep source layers identifiable, and validate the model before it carries an analytical or project commitment.
Top comments (0)