Separate a file error from a display difference
Start with the Khronos glTF Validator and record its errors and warnings. Its browser version processes the selected files locally. A clean validation result means the checked rules passed; it does not prove that the asset looks correct, loads quickly or meets a particular platform's requirements.
Open the same export in the Labs GLB viewer, then the target website or app. Keep a reference image from the source application. Differences between those three views help locate whether the issue belongs to export, interpretation or the destination configuration.
Check material inputs individually
A native shader graph is not automatically portable to every glTF viewer. Check what the exporter supports, then inspect the exported material rather than assuming each source node survived.
In the glTF metallic-roughness model, base-colour texture RGB uses sRGB encoding, while roughness and metallic data are linear. Packed metallic-roughness textures use the green channel for roughness and blue for metallic. Inspect the assigned maps and factors before changing the lighting to compensate.
For missing textures, verify that the required images are embedded or remain available at their referenced locations. A GLB container can still reference external resources; the extension alone is not proof that the handoff is self-contained.
Check geometry, orientation and scale
- Compare a known product dimension with the export and the intended real-world presentation; do not infer scale from how the viewer frames the model.
- Turn the model through its full range and inspect back faces, normals and transparent areas.
- Check the intended UV set, normal map and material assignment on a close view.
- Confirm which extensions and compression formats the receiving viewer supports before making them part of the delivery.
Test the delivered experience
Measure loading and interaction on a representative phone and connection, using the actual asset URL. Browser GPU memory and texture decoding matter alongside downloaded bytes. Keep the benchmark guide's planning targets separate from destination-specific requirements.
For AR, test the intended route on the actual device. Browser-side material or transform changes do not necessarily carry into an external AR viewer. Record the tested device, browser, route and asset version in the handoff so the next person can reproduce the result.