Imported 3D models at the wrong scale or position: checking units, origins and axes

Separate the three symptoms before changing settings

A model that is uniformly tiny calls for a dimension check. A correctly sized model lying on its side calls for an orientation check. A model with the right shape and orientation but a distant location calls for a reference-position check. Several errors can coexist, but changing one variable at a time makes the diagnosis traceable. This guide concerns visualization assets, not engineering surveying or geospatial calibration.

Choose a dimension supported by the source material, such as a documented cabinet height, and identify its endpoints. “About the height of a person” is not a reliable scale reference. A similar-looking screenshot is not a measurement either. When the reference itself is disputed, the information owner must confirm it before the modeler treats it as a real dimension.

Record four checkpoints in the conversion chain

1. Source file: note the authoring application, file version, length unit, reference dimension and object scale. Establish whether the visible size comes from the geometry or an additional object transform.
2. Export: record the format, exporter version, unit conversion and up-axis setting. Test a copy and retain the untouched original.
3. Import: record the receiving application, scale factor, axis conversion and any automatic centering. A camera that automatically frames the object does not establish correct dimensions.
4. Assembly: inspect the asset alone, then under its intended parent and alongside other assets. Identify the first stage at which the error appears.

File conventions matter. The Khronos glTF 2.0 specification defines meters for length and positive Y as up. This statement applies to glTF 2.0; it does not establish the units of an arbitrary CAD file. The specification was checked on October 6, 2026.

Placement needs a shared reference

When a building and its equipment are exported separately, agree on a common assembly reference rather than centering every asset independently. A traceable intersection on an approved plan can serve as a visualization reference, with orientation and relative positions recorded. Keep a relationship between the original-coordinate material and the transformed visualization assets.

Hypothetical example: a building uses its southwest corner as zero, while an equipment file uses the center of the machine. Both appear correct alone but overlap incorrectly when assembled. The missing information is the placement relationship. Stretching the building to fit the machine would conceal the problem. This example illustrates a reference mismatch, not a real project measurement.

Evidence that the correction is reproducible

• Measure the same reference edge in the receiving application. Its value should match the approved input within a tolerance agreed for the intended use.
• Check the front, top and an asymmetrical feature. A mirrored model may otherwise look plausible.
• Align the asset with its neighbors, export and import again, and confirm that placement survives without undocumented manual dragging.
• Deliver the source and target units, reference point, axis convention, transformation values and verification views. Another person should be able to repeat the process.

If only one child part has the wrong size, investigate that part and its parent rather than scaling the complete scene again. Likewise, alignment inside software does not prove that uncertain field information is spatially accurate. Keep the limits of the original evidence visible in the acceptance record.

Can work continue without a known dimension?

A clearly identified visualization mock-up can support discussion of shape and composition, but dimension acceptance remains unresolved. Before joining it to existing assets, obtain at least an agreed set of dimensions and placement references. Otherwise the next import can invalidate a previous manual correction, making an apparently finished file expensive to maintain.

Attach dimension, origin and axis references to a sample before deciding whether existing assets can be reused. Once placement is consistent, assess whether optimization changes the appearance. Provide the receiving environment and agreed reference object when discussing 3D modeling services (Chinese); these inputs make repeated manual scaling and repositioning less likely to become the handover method.