I wanted a workflow that could answer two different questions without mixing their outputs. The first question is about an object: what might a building, pavilion, vehicle, or street element look like? The second is about a place: how does a proposed volume relate to roads, terrain, and nearby buildings?
Meshy and Shapezo fit those questions differently. Meshy uses AI and parameter-driven controls to generate stylized 3D assets from prompts or references. Shapezo uses a map-first process: I outline a study area, and its AI generates a starting model for the selected geography. This article shows how I keep the two stages separate and how I validate the handoff.
Step 1: Define the output before opening a tool
I write one sentence for the decision I need to support. “Compare three facade directions” points to Meshy. “Check whether a new block connects to the existing street network” points to Shapezo. If the sentence contains both an object and a site, I split it into two smaller tests. That keeps a fast visual study from being mistaken for a complete urban model.
Step 2: Generate the object study in Meshy
For an object study, I collect a short prompt, one or two references, and a target scale. I ask Meshy for a few clearly different options rather than many minor variations. I inspect silhouette, openings, roof geometry, material cues, and the amount of detail that survives at the intended camera distance.
I then record the prompt, reference names, export format, and any manual edits. The generated mesh may need normal fixes, topology cleanup, decimation, UV work, or a complete rebuild of structural elements. Those are normal steps. The model is an exploration artifact, not evidence that dimensions or construction logic are correct.
Step 3: Generate the site study in Shapezo
For the site study, I draw a boundary in Shapezo that includes the project parcel and enough surrounding context to explain access and scale. I generate the initial AI model, then check location, orientation, approximate building heights, street continuity, open space, and visible terrain breaks.
I save the selected boundary and the data date. I also create a short assumptions list: which roads appear simplified, which heights look estimated, and which features may be missing. That list is more useful than a polished screenshot because it tells another person what still needs verification.
Step 4: Test the handoff
I export a simplified Meshy asset and place it into a conventional scene or a visual study beside the Shapezo context. I align units, origin, and orientation before judging the composition. If the two coordinate systems do not match, I do not hide the mismatch with camera tricks. I fix the transform or mark the result as a visual-only overlay.
I compare the same views for every option: aerial, street-level, and oblique. I check frontage, height, shadow direction, pedestrian approach, service access, and the relationship to parks or water. I keep the camera settings unchanged so the comparison is about the design and context, not about a more flattering angle.
Step 5: Validate what can be trusted
I classify each element as observed, imported, generated, estimated, or manually edited. For Meshy, that status usually applies to the object geometry and material cues. For Shapezo, it often applies to roads, buildings, terrain, and parcel context. I do not use either generated result as a survey, permit drawing, code check, or construction document.
When a direction is worth more effort, I rebuild the important geometry in a controlled tool such as Blender, Revit, SketchUp, or a civil platform. I keep the early models as references, not as unquestioned source files. This makes the transition slower in the right places and prevents a convincing approximation from quietly becoming a project fact.
Common mistakes I try to avoid
The first is polishing an asset before checking its scale in context. The second is using a generated street or building as if it came from a current survey. The third is comparing options from different camera positions. I keep a short decision log with the chosen view, the source date, and the next verification step. It takes a few minutes and saves me from arguing later about whether a change came from the design or from the model setup.
The repeatable rule
Meshy is my fast object generator. Shapezo is my fast place generator. I get the best results when I define the question first, generate only the geometry needed for that question, and preserve a small record of assumptions. The workflow stays light at the beginning, but the limits of AI-generated geometry remain visible when decisions become more serious.
Top comments (0)