I get better results when I separate concept generation from city-model assembly. Shapezo is useful for the first stage: select an area on a map and generate a rough 3D proposal with AI. PlaceMaker is useful for the second: bring CAD and map data into CityEngine and build a structured city context with buildings, roads, terrain, parks, water, and infrastructure.
The key difference is data status. Shapezo produces inferred geometry for exploration. PlaceMaker works from source layers and procedural or parameterized rules. A reliable pipeline keeps those states visible so a fast concept does not get mistaken for verified city data.
Stage 1: define a bounded experiment
I start with a decision and a boundary. A station block, campus edge, industrial parcel, or waterfront segment is easier to evaluate than an undefined request to model an entire city. In Shapezo, I use the boundary to test massing, access, open space, and broad program relationships. I save the boundary, prompt assumptions, and date with each concept.
At this stage, I do not expect reliable object semantics. The AI may suggest a plausible building or street without knowing the exact survey, right of way, utility corridor, or design standard. That is acceptable for a hypothesis, as long as the asset is labeled concept-only.
Stage 2: assemble source layers
When an option deserves more work, I prepare the CityEngine context through PlaceMaker. The available inputs can include DWG, DXF, KML, GeoTIFF, Shapefile, PDF, or SVG, depending on the project. I check coordinate reference, units, coverage, layer meaning, and update dates before bringing the data into the scene.
PlaceMaker is useful because it treats urban content as more than a set of surfaces. Buildings, roads, green areas, water, and terrain can be generated or organized as city elements. That gives me a better foundation for later rules, visualization, analysis, and exports such as CityGML, OBJ, PLY, STL, or 3D Tiles.
Stage 3: refine the selected concept
I do not need to replace every rough object at once. I prioritize the geometry that affects the current decision. For an access study, that may be roads, entrances, and terrain. For a view study, it may be building height and surrounding blocks. For a public-space review, it may be paths, plazas, landscape, and water edges.
The city context also gives me a place to apply consistent parameters. Building height, floor count, vegetation density, road width, and other rules can be adjusted across a district instead of edited one object at a time. The exact parameters depend on the source and purpose, so I record them rather than assuming a default is always correct.
Stage 4: validate before analysis
A CityEngine scene can support daylight, wind, viewshed, movement, land-use, or energy questions, but the scene is not automatically a valid analysis model. I check scale, missing layers, terrain quality, object completeness, and the date of the underlying data. A model built for public visualization may need a different level of detail from one built for simulation.
Failure modes I watch for
The first failure mode is confusing a selected map boundary with a complete site description. The second is treating AI massing as if it contains survey or engineering truth. The third is exporting without checking coordinate systems and destination requirements. I avoid these by maintaining explicit model states: generated concept, structured city context, and validated analysis input.
I also keep a small layer manifest with the project. It records which source produced each major object, whether a rule generated it, and what level of detail it carries. When a reviewer asks why a building, road, or water surface looks different from the source, I can trace the change instead of guessing. That makes CityEngine scene updates much easier to repeat.
My implementation rule
Use Shapezo to decide what deserves modeling effort. Use PlaceMaker and CityEngine to build the connected context that lets the decision be reviewed. Keep the handoff traceable, keep source layers identifiable, and validate any model before it carries an analytical or project commitment.
