Why won't my 3D scanner pick up black or shiny parts?

Because an optical scanner measures light that comes back to it diffusely, and those two surfaces break that in opposite ways. A glossy or machined face reflects the projected pattern away from the camera like a mirror, so the sensor sees either nothing or a blown-out highlight. Matte black absorbs most of the light that lands on it, so there is too little signal to triangulate. Transparent parts fail a third way — the light goes through rather than coming back. All three are fixed by making the surface temporarily matte and opaque, which is what scanning spray is for, not by rescanning more slowly.

Does scanning spray change the accuracy of the scan?

It adds thickness, so it can, and how much depends entirely on how it is applied. A light, even pass from 20–30cm adds a coat that manufacturers of self-vanishing sprays state in microns — negligible against the 0.05–0.1mm accuracy of a consumer handheld. A heavy, wet, uneven coat is a different matter: it builds unevenly in corners and pools on horizontal faces, and that error goes straight into your mesh as real geometry. Spray in short passes from a distance rather than close and heavy, and apply it to a clean, dry part.

Why is my 3D scan bent or banana-shaped at one end?

That is accumulated registration drift, and it is a long-part problem. The scanner aligns each new frame to the previous ones, and every alignment carries a small error. On a short part the errors stay small. Along a metre of extrusion or a bumper, each frame inherits the last frame's error and the deviation compounds until the far end is visibly off. The fix is not a slower pass — it is fixed references that do not accumulate: positioning markers along the part, or scanning in overlapping sections and aligning each to features you can measure independently.

Why does my scan show two copies of the part at an angle to each other?

The scanner lost tracking and then re-registered the incoming frames onto the wrong place. It usually happens over a featureless region — a flat plate, a plain cylinder, a large blank face — where geometry tracking has nothing unique to lock onto, or when the scanner moved faster than the software could follow. Stop, delete the orphaned section rather than trying to fix it in the mesh, and rescan that region with markers or with the part repositioned so a feature-rich area stays in view. A doubled mesh is a capture failure; no amount of cleanup recovers the true surface from it.

Can I scan a part and 3D print the copy directly?

For a shape — a sculpture, a prop, a cosmetic panel — yes, clean up the mesh and print it. For a functional part, usually not. Scan data is a surface capture with noise and tolerance baked into it, so holes come out slightly oval, flat faces are slightly wavy and nothing is exactly nominal. Printing that mesh gives you a copy of the measurement, not a copy of the part. For anything that has to fit, mate or seal, rebuild the geometry in CAD using the scan as reference, which is the reverse-engineering workflow rather than the copy workflow.

When is a scanning service cheaper than fixing my own scan?

When the part is large, highly reflective, metrology-critical, or due to someone tomorrow. Chasing a bad capture costs an evening per attempt, and consumer hardware has a genuine ceiling on long parts, deep internal features and tight tolerances that no technique clears. A shop with a tracked or standalone structured-light system and a calibrated process gets one clean capture and a usable mesh. The directory at /providers lists providers by location and process, and a single quote often costs less than the third failed weekend.

Why 3D Scans Come Out Noisy or Full of Holes

The mesh came back and half the part is missing. There are holes straight through the machined faces, the surface that did capture has a fuzzy texture that the real part does not have, and somewhere along the length the whole thing has quietly bent. A bad 3D scan is not random, though — each of those defects has one dominant cause, and most of them are about the surface and the tracking rather than the scanner. This guide reads the symptom first and works the fixes from free to expensive, ending at the point where the honest answer is having the part scanned by a service.

This post contains affiliate links. If you purchase through these links, 3D Prototyping Hub may earn a small commission at no extra cost to you.

This post also contains Amazon affiliate links. As an Amazon Associate we earn from qualifying purchases.

Read The Defect Before You Change Anything

Four failures account for nearly every disappointing capture, and they do not share a fix. Match yours before touching a setting.

What you see	What it actually is	Where to start
Holes over shiny, black or clear areas	No usable light returned to the sensor	Matte the surface
Fuzzy, grainy surface everywhere	Wrong working distance, exposure or ambient light	Distance and lighting
Two copies of the part, at an angle	Tracking lost, then re-registered wrongly	Markers, rescan the region
Long part bent or twisted at one end	Accumulated registration drift	Fixed references, sectioned scans
Small features simply absent	Point spacing coarser than the feature	Measure them, rebuild in CAD

Holes Appear Where The Surface Won't Cooperate

An optical scanner projects a pattern — structured light or a laser line — and watches how that pattern deforms from a camera mounted at a known offset. That geometry only works if the surface scatters light back diffusely. Three surfaces break it.

Glossy and machined metal reflect the pattern away like a mirror, so the camera sees a blown-out highlight in one spot and nothing anywhere else. Matte black absorbs nearly all of it, leaving too little signal to triangulate. Clear and translucent parts let the light through and refract it, which produces data that is present but wrong — the worst of the three, because it does not look like a failure.

The fix is to make the surface temporarily matte and opaque. In order of preference:

A self-vanishing scanning spray. A light, even pass gives a uniform matte coat that sublimates away over a few hours, leaving nothing to clean off. This is the standard answer and it is worth the price the first time it saves a rescan — anti-glare scanning sprays are sold in aerosols small enough for occasional use.
Talc, chalk or developer powder as a stop-gap. They work, they are cheap, and they need cleaning off a finished part. On a decorative object that is fine; on a part going back into a machine, less so.
Changing the angle before either. A specular face that returns nothing at 90° often returns clean data at 30°. Free, and worth ten seconds before reaching for the can.

Spray from 20–30cm in short passes. A heavy wet coat pools in corners and on horizontal faces, and that pooling is measured as geometry — you have swapped a hole for a lie.

Fuzz And Noise Usually Mean Distance Or Light

If the whole surface has a grainy crust a millimetre thick, the capture itself was marginal. Three causes, in the order worth checking.

Working distance. Every scanner has a stated working range and a depth of field inside it. Data thins and roughens fast outside that band, and handheld operators drift out of it constantly — especially while walking around a part. If your scanner shows a live distance indicator, work to it rather than to what feels comfortable. For small objects, a motorised turntable removes the problem entirely by holding the part at a fixed distance and letting it come to you.

Ambient light. Direct sun is the reliable killer: it is broadband, it is bright, and it swamps the projected pattern. Overhead fluorescents and IR-rich halogens interfere too. Scan indoors, away from windows, under stable diffuse light — and do not change the lighting halfway through, because the exposure calibration you set at the start no longer applies.

Exposure and calibration. Most scanners calibrate against a supplied board, and most people do it once. Recalibrate after transport, after a temperature change, and any time the noise floor climbs for no other reason. It takes two minutes.

Doubled And Bent Meshes Are Tracking Failures

This is the defect people spend the longest fighting in software, and it cannot be fixed in software.

A handheld scanner has to work out where it is for every frame. It does that by matching the incoming frame to what it has already captured, using geometry, markers, or texture depending on the mode. Geometry tracking needs distinctive shape in view. Point it at a flat plate, a plain cylinder, a pipe or a large blank panel and there is nothing unique to match — so it guesses, and when it guesses badly you get a second copy of the part floating at an angle to the first.

Drift is the same mechanism playing out slowly. Every alignment carries a small error, and each new frame inherits the last one's. Across a short bracket that stays invisible. Along a metre of extrusion the errors compound until the far end is out by more than your tolerance, which is the classic banana-shaped scan.

The fix for both is references that do not accumulate:

Positioning marker dots on the part and on the surface around it. Place them irregularly — an even grid gives the software repeating patterns to confuse. The mat in the photograph at the top of this page is exactly this, done as a fixture.
Overlapping passes. Each pass should share a generous band with the last, and the overlap should contain features or markers, not blank surface.
Sectioned scans on long parts, each aligned to features you can check independently with calipers, rather than one heroic single pass.

And when tracking is lost, delete the orphaned section instead of stitching it. A mesh containing two versions of the same face has no correct answer hiding inside it.

Some Features Were Never Going To Be Captured

Two hardware limits get mistaken for user error, and knowing them saves whole evenings.

Point spacing is not accuracy. A scanner specified at 0.05mm accuracy may sample points 0.1–0.2mm apart. Any feature smaller than that spacing — a 0.3mm engraved line, a sharp edge break, a fine knurl — is not going to resolve however many passes you make, because the scanner never sampled it.

Optical scanning is line of sight. A deep blind hole, an internal channel, a tight undercut: if the camera cannot see the surface, it does not exist in the data. No setting changes this. Measure those features with calipers, pin gauges or a bore gauge and put them into the CAD model by hand, which is faster and more accurate than any amount of rescanning.

This is also the honest boundary of mesh repair. Automatic hole-filling interpolates a plausible surface across a gap — useful on a cosmetic object, and quietly wrong on a sealing face or a bore. If the part has to fit something, rebuild the geometry in CAD from the scan rather than printing the mesh; the reverse-engineering workflow exists precisely because a mesh is a measurement, not a model.

When The Scanner Is Genuinely The Ceiling

Technique fixes most bad scans. Some it does not, and the tell is consistency: clean captures on small feature-rich objects, persistent failure on large, dark, or tight-tolerance work. What is worth knowing before spending, stated from specifications rather than from a shootout we did not run:

Consumer handhelds are generally specified around 0.05–0.1mm accuracy, with blue-light models rated tighter on small objects — enough for reverse engineering most mechanical parts and for capture work. Revopoint's handheld scanner line sits in this band, and the current models list their accuracy and scan volume per unit — read those two numbers against your actual parts, not against the headline.
Standalone structured-light systems are the step up when scanning is a production function: better behaviour on difficult surfaces, stronger software for clean CAD export, and repeatability you can put in a report. SHINING 3D's EinScan standalone range starts well into four figures and is bought for exactly that reason.

If you are choosing rather than diagnosing, the full comparison is in our 3D scanner buyer guide, which covers scan volume and workflow alongside accuracy.

When To Stop And Hand It Over

Three signals that you are past the point of diminishing returns: the part is large enough that drift dominates, the tolerance is tight enough that the scanner's specified accuracy is the same order as the feature you care about, or the scan is blocking someone else's work. Any of those, and one professional capture is cheaper than another weekend.

That is also true when the scan is an inspection job rather than a modelling job — comparing a manufactured part against nominal CAD is a measurement task with a report attached, and it usually needs metrology-grade equipment to mean anything.

For everything else, work the list in order: matte the surface, fix the distance and the light, give the tracker something to hold onto, and accept the features the hardware was never going to see. Most "bad scanner" problems are one of those four. When they are not, the providers in our directory scan for a living and will send you a mesh you can build on.

Hero photo by Creative Tools from Halmdstad, Sweden via Wikimedia Commons, licensed CC BY 2.0.