I approach a closed retail site as a systems problem. There is a building shell, a parcel boundary, a street network, utilities, parking, service access, and a list of community needs. If I change one part, the others react. A new housing wing changes fire access. A clinic changes privacy and deliveries. A plaza changes stormwater and pedestrian crossings. My goal is to make those dependencies visible before detailed design begins.
Step 1: Capture the Existing Constraints
I start with a short site record. I note the building footprint, floor to floor height, loading docks, roof equipment, major entrances, parking aisles, bus stops, sidewalks, trees, and nearby homes. I also mark unknowns rather than inventing certainty: slab capacity, hazardous materials, utility condition, easements, and exact grades need field verification.
Then I list program candidates in plain language. For a former big box store, that might be a public market, a maker space, a sports center, a clinic, a library, childcare, and housing. Each candidate gets a few non negotiables: daylight, acoustic separation, service route, operating hours, and emergency egress.
Step 2: Generate Comparable Massing
This is where I use Shapezo. I select the target area on a map, and the AI generates an initial 3D model with site context and a rough building mass. I do not ask it for a finished building. I ask for a neutral baseline that lets me compare options.
For the same parcel, I might create three schemes: a market with a sports center, a clinic and library hub with housing above, and an indoor street with small storefronts. I keep the prompt focused on geometry and relationships. The useful output is not a polished facade; it is the location of entrances, courtyards, service lanes, and walking routes.
Step 3: Run Checks That an Image Cannot Run
An AI massing model can suggest a courtyard, but it cannot certify that a wheelchair user can reach every public room. I translate the images into explicit checks:
Access and Movement
I trace a continuous accessible route from the public sidewalk to each shared use, housing lobby, restroom, and outdoor space. I separate deliveries from play areas and verify that a bus or fire truck can turn without crossing the main plaza.
Building Performance
I ask where daylight can reach deep interiors, where mechanical rooms could connect to existing shafts, and whether a new upper floor creates an unreasonable structural load. These are questions for engineers and code specialists, not for the image generator.
Operations
I map opening hours and secure boundaries. A library might stay open after the clinic closes. Childcare needs controlled entry. A gym may need evening access. The plan should allow each use to operate without unlocking the entire building.
Step 4: Test the Outdoor Retrofit
The parking field deserves its own iteration. I test a baseline with existing parking, a phased option with trees and temporary markets, and a final option with a plaza, housing, bike storage, and stormwater gardens. I keep accessible parking and loading close to doors, then measure the walking distance between the bus stop and the most public entrance.
What I Trust, and What I Do Not
I trust AI generated massing to accelerate comparison and expose spatial questions. I do not trust it to infer legal setbacks, existing structural conditions, utility capacity, insurance requirements, or community consent. I also check for visual errors: impossible stairs, rooms without windows, duplicated doors, and roads that terminate at walls.
The workflow works because it keeps the model provisional. I can show a clear set of alternatives, record assumptions, and invite criticism while change is still affordable. The final design still belongs to qualified professionals and the people who will use the place. AI helps me move the first conversation forward; it does not remove the responsibility to verify every important decision.
Top comments (0)