What One Image Cannot Tell You About a 3D Object

Single-image 3D tools are useful, but every result contains assumptions.

Hidden surfaces are inferred

The back, underside, and occluded joints are not visible. A system can mirror the front, use category priors, or ask for another view, but it cannot measure pixels that do not exist.

Scale is ambiguous

Without a known reference, a small toy and a full-size object can project to the same image. The model can preserve proportions while still getting real-world scale wrong.

Lighting is mixed with material

A dark pixel may be paint, shadow, dirt, or an inward-facing surface. Reliable reconstruction separates albedo, roughness, normal response, and illumination instead of copying color into every material channel.

Review loops matter

For procedural Three.js work, the most valuable step is a side-by-side render review. Fix the sculpt spec when the structure is wrong; fix the code when the specification is sound but the render misses the reference.