The work behind a manufacturing decision

Reference material can be drawings, a layout export, screenshots, photographs, or an image-to-3D mesh. That material is interpreted into parametric CAD, passed through geometry validation, handed to the slicer, checked for physical fit, and then taken into production. The shape is straightforward. The value is in retaining each interpretation and decision as the project moves along it.

The saving is not that a machine writes the CAD. It is that the interpretation, validation and slicer setup are recorded and repeatable. When a correction appears on the bench, that correction feeds back into generation instead of being redone by hand from disconnected notes and remembered settings.

Validation is a gate, not a formality. Every printable STL is checked for one intended connected shell, zero non-manifold edges unless they are explicitly accepted as a known mesh-source artifact, on-plate Z placement, and bed fit. A file that looks complete in a viewport can still fail one of those tests. The record makes the result visible before the part is treated as ready for production.

Image-to-3D gets a fast, useful base. It does not get a printable part. The building tower began as a watertight, single-component image-to-3D mesh of about 2.0 million triangles. That gave the rebuild a visual starting point without requiring every facade mass to be reconstructed from nothing. The output still needed to become a parametric design before it could carry the next decisions.

The transit coach model shows the same boundary from imported SKP meshes. They brought feature density and the rhythm of its livery, doors and windows into the rebuild. But imported meshes do not carry semantic intent. A window is not known to be a window, facade planes are not reliably flat, and balconies, cornices and mullions confuse automated selectors. Openings have to be inferred from depth and coordinate probes. A small topology edit can produce holes or a non-manifold explosion. That limit is precisely why the rebuild and validation exist.

A resin-intended STL is checked for build-volume fit, connected shells, non-manifold and boundary edges, hollowing state, drain and vent requirements, thin features and orientation candidates. It is then sliced in a resin slicer and the sliced output is checked with UVtools. That step reads the result as a resin job, including what the supports and exposure result actually contain.

Islands are triaged into pixel noise, groove or seam slivers that may be fixable in the bitmap, and genuinely unsupported regions that need resupport or reorientation. A human decides whether to print, repair or regenerate. The helmet established this runbook, and the console replica and sports car kit follow its discipline. The distinction matters because a resin file can be geometrically plausible yet still need a different support or orientation decision after slice inspection.

Read how geometry and route decisions are made. Compare the two processes.