You scanned the part, the mesh on screen looks exactly like the part, and then everything stops. The slicer loads the file and shows an empty plate. Or it slices and the preview is a nest of stray walls with no inside. Or it prints, and what comes off the plate is a lumpy, open-sided suggestion of something you recognise. When a 3D scan won't print, the printer is almost never the problem, and pressing Repair is almost never the fix.
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 I earn from qualifying purchases.
A scanner and a slicer want two different things from geometry. A scanner records a surface it observed. A slicer needs a closed volume it can fill. Those are not the same object, and the gap between them is specific, small, and worth understanding once.
Read What Stopped, Before You Repair Anything
Five failures account for nearly all of this, and they look different enough to tell apart in under a minute.
| What you see |
What it actually is |
| Slicer loads the file and the plate is empty |
Mesh is not closed, or normals are flipped |
| Preview shows loose walls, no infill, hollow layers |
Open edges — the slicer cannot trace a closed loop |
| Model is microscopic or the size of a building |
Unit mismatch between exporter and importer |
| Slicer hangs, crashes, or takes forever |
Triangle count, not geometry |
| It prints, but soft and vague where the detail was |
Printer resolution, or texture mistaken for geometry |
Diagnose first. Every one of these has a different fix, and three of them are made permanently worse by an automatic repair pass.
A Scan Is a Shell. A Print Needs a Solid.
A mesh is a bag of triangles describing a surface. For a slicer to use it, that surface has to enclose a volume without ambiguity: every edge shared by exactly two triangles, every normal pointing the same way out, no triangles passing through each other.
Slicing is a sectioning operation. At each layer height the slicer intersects the model with a horizontal plane and traces the closed loops where they meet — inside the loop is plastic, outside is air. An edge that belongs to one triangle leaves a loop that never closes. An edge shared by three leaves a loop that branches. Flip the normals and the slicer concludes the solid is everything outside your part, which is why an inside-out mesh sometimes produces a block with a part-shaped hole in it.
This is why the symptom is so binary. The file is not slightly wrong; the question "is this point inside the model" simply has no answer.
Holes Are Missing Data, Not Missing Triangles
A hole in a scan is a place the sensor never saw: a bore it could not look down, an undercut, the face the part was sitting on, or a patch that was too shiny, too dark or too translucent to return a reading. Our piece on scans that come out noisy or full of holes covers why the surface defeats the sensor in the first place.
What matters here is what happens next. Automatic hole-filling spans the gap with a membrane that is geometrically closed and dimensionally invented. The file now slices. Nobody is told that a surface was fabricated, or by how much it missed.
So split the holes into two piles:
- Holes you can afford to invent. A gap on a smooth, unimportant region — a flank, a decorative face, anywhere nothing mates. Patch it, match the curvature, move on.
- Holes you cannot. Bores, seating faces, mounting pads, anything that touches another part. Re-scan those. Flip the part, capture a second orientation, merge the two.
The seating face is the one that catches people. It is missing from essentially every single-orientation scan, it is flat and therefore trivially easy for a repair tool to close convincingly, and it is very often the datum the printed copy has to sit on.
The Scan Has No Thickness
Scan one side of a panel, a body shell, a mould face or a cover, and the mesh you get is an open surface — a sheet in space with zero volume. There is nothing to print, in the most literal sense available.
The fix is to give it a wall. Offset the surface to a real thickness before you go anywhere near a slicer: 1.5–2mm is sensible for a cosmetic FDM copy, 2.5–3mm if the part will be handled or take any load. Resin needs the same treatment plus drain holes if you hollow it, or the vat pressure will dimple the walls on the way up.
Decide the thickness deliberately rather than accepting a default. The default is usually 1mm, which prints as two perimeters and nothing between them, and snaps when you pick it up.
Too Many Triangles, and What the Slicer Does With Them
Raw captures of two to ten million triangles are routine. That density exists to describe measured surface at the scanner's sampling spacing, and no FDM or resin printer can reproduce detail at that spacing — you are asking the slicer to section ten million facets to produce a 0.2mm layer that could not express the difference.
Crop first: delete the mat, the fixture, the table, the operator's hand. Then decimate to somewhere between 200,000 and 500,000 triangles. The surface error that sensible decimation introduces sits comfortably below the scanner's own accuracy, so nothing real is lost, and the file stops fighting every tool it touches.
Order matters. Crop, then repair, then decimate, then scale. Decimating a mesh that still has open edges tends to convert small holes into larger and stranger ones.
The Scan Is the Wrong Size and Nothing Warns You
STL and OBJ carry coordinates and no units. The exporter wrote numbers; the importer assumed millimetres, or inches, or metres. If those assumptions disagree you get a part 25.4 times too large or a thousand times too small, and no software reports an error, because the file is not malformed — just unlabelled.
Photogrammetry has a stronger version of the same problem: reconstructing geometry from photographs recovers shape only up to an unknown scale factor, so the result has no inherent size at all.
The fix is physical, and it takes a minute. Measure one real feature on the real part with digital calipers — the longest one you can reach cleanly, because a percentage error shows up most on the longest dimension. Scale the mesh so that feature matches. Then measure a second, independent feature and check it. If both agree, the whole model is right; if they disagree, the capture itself has drift and no uniform scale factor will rescue it. Tolerances and accuracy in 3D printing is worth reading before deciding how close is close enough.
Detail You Can See and the Printer Cannot Make
Two different things get mistaken for each other here.
The first is texture. Most scanning software drapes a high-resolution photographic texture over a much coarser mesh, and the texture carries the apparent crispness — lettering, grain, fastener heads. Switch to a shaded, untextured view before you judge any scan. What you see there is what will print.
The second is genuine resolution. A 0.4mm nozzle cannot render a 0.25mm scanned engraving; it arrives as the soft mush described in why printed text and logos come out mushy. Resin is a different proposition and reproduces scanned surface detail far better — if the detail is the point of the copy, print it on resin and accept the workflow, which the current resin machines handle without drama.
Resin is also unforgiving in the other direction: scanner noise at the 0.05mm scale is invisible on an FDM print and arrives as visible orange-peel on a resin one. Smooth the mesh lightly before printing fine work.
Fix It at the Capture, Not in the Repair Tool
Every repair is an invention. Every minute spent capturing properly removes one.
Matte the surfaces that return nothing. Shine, dark colours and translucency are the cause of most holes, and they cluster on exactly the machined faces and bores you care about. A light pass of self-vanishing scanning spray converts an invented surface into a measured one, and sublimates away on its own.
Capture two orientations, always. The underside is not optional on a part that has to sit flat. Scan it upright, flip it, scan again, merge. This single habit closes more meshes than any repair algorithm.
Keep the distance constant. For anything bench-sized a motorised turntable gives even coverage and consistent overlap, which is most of what registration needs. Hand-walking a scanner around an object is how coverage gaps happen.
Use markers on anything long. Handhelds accumulate alignment error frame to frame, so a 500mm part can come back gently bowed while every local detail looks perfect — and a printed copy of a bowed scan will not drop into the assembly it came out of. Positioning marker dots on the fixture and surrounding surfaces give the software fixed references. Keep them off features you need.
Match the scanner to the job, which for printing is shape rather than metrology. Revopoint's handheld line covers bench-sized objects well; the figure to read per model is point spacing, because that is what decides whether a feature arrives as geometry or as noise. At the budget end, 3DMakerpro's handheld scanners are honest shape-capture tools, and shape capture is a far more forgiving job than reverse engineering. Where scanning is a recurring function, SHINING 3D's EinScan standalone range is the fixed-rig answer: a calibrated working volume returns a consistent, closed surface across a whole part rather than one assembled from drift.
A Sequence That Ends in a Printable File
- Open the scan untextured and look at the actual surface.
- Crop away the mat, the fixture and everything that is not the part.
- List the holes. Decide which may be invented and which must be re-scanned.
- Re-scan the ones that must be, in a second orientation, and merge.
- Fill the remainder deliberately, matching curvature rather than flat-patching.
- Close the mesh: open edges, flipped normals, self-intersections, isolated geometry.
- Thicken, if the capture is an open surface.
- Decimate to 200,000–500,000 triangles.
- Scale from one calipered feature, then verify against a second.
- Slice, and check the first and last layers in preview before committing four hours to it.
Step 10 costs ten seconds and catches the inverted normal that steps 1 through 9 missed.
When to Hand It Over
Three cases where sending it out is simply cheaper. The object is larger than anything you own can capture in one piece or print without splitting. The copy has to fit an existing assembly, which makes it a dimensional job rather than a shape job. Or the part should be modelled rather than copied — a prismatic bracket is faster and more accurate rebuilt from measurements than reproduced from a mesh, which is the argument in why 3D scans don't work in CAD.
Ask for the right deliverable. "Scan and print this" gets you a copy; "a watertight mesh scaled from physical measurements, with the seating face captured rather than reconstructed" gets you a copy that fits. Our guide to 3D scanning services for reverse engineering covers how those quotes are structured, STL vs STEP files is five minutes well spent before you specify anything, and the provider directory lists shops by location and process so the part does not have to travel far.
A scan is a record of a surface. A print needs a solid. Most of the work is converting one into the other honestly — knowing which millimetre was measured and which was filled in.
Hero photo by Creative Tools from Halmdstad, Sweden via Wikimedia Commons (CC BY 2.0).
