Doubts Worth Raising

Usually not. AI can propose candidates constrained by surviving geometry and comparisons. Multiple historical forms can fit the same fragment, so an attractive completion is not proof.

An archival image or a matching surviving piece may narrow the possibilities considerably. The strength of the conclusion comes from that evidence, not from the software's fluency. Explore reconstruction methods.

No. Manual modelling, photogrammetry, geometric fitting and mirroring can support reconstruction without generative AI. Each method has different assumptions.

Photogrammetry primarily estimates a surface from photographs. A completion model proposes missing information. Keeping these tasks separate makes it easier to understand which regions are measured and which are inferred.

No. Resolution concerns recorded detail; accuracy concerns agreement with the object. Calibration, scale, material behaviour and independent checks remain important.

A dense point cloud can still contain systematic distortions. Ask how scale was established and which checks were made outside the processing workflow. Read about surface records.

It can inform assessment, but does not authorise treatment. Conservators must separately consider material compatibility, significance, stability and future care.

A proposal suitable for a screen might be unsuitable as a printed insert or a permanent fill. Digital plausibility does not resolve adhesives, mechanical stress or the consequences of handling an original object.

Alternatives reveal when the evidence supports several possibilities. Showing them avoids presenting a preferred hypothesis as the only historical answer.

Alternatives need not overwhelm a display. A second profile, a labelled uncertainty layer or a short explanation of the unresolved region can make the distinction clear.

Its meaning depends on the model and evaluation. A similarity score is not automatically the probability that a reconstruction is historically correct.

Before accepting a percentage, ask what was measured, how the score was tested and whether the test data resemble the object being studied. Explore validation.

A photograph alone generally cannot establish pigment composition. Lighting, ageing and camera processing affect colour; material identification requires suitable analysis.

Imaging and analytical methods can complement one another, but an AI-generated colour restoration should not be described as recovered original pigment without supporting evidence.

Public visibility is not blanket permission. Check item-level licences, terms, applicable law and any cultural restrictions before reuse.

Rights in the photograph may differ from rights in the depicted object. Cultural sensitivity may also call for consultation even where a file is legally reusable. Read about rights and disclosure.

Look for labels identifying surviving material, proposed additions, sources and unresolved regions, together with a date or version and an explanation of method.

If a display does not provide these, a useful question is: “Which parts were recorded from the object, and which parts were proposed?” That distinction is a good starting point for any digital conservation story.