STL Repair Tool

Yes. Repairing, merging, scaling, simplifying and downloading are free, with no sign-up, no watermark and no limit on how many files you fix. The only limits are your computer’s memory and the file size (see below).

No. Your file is read and repaired by your own browser, on your own device, in a background thread. Nothing is sent to a server – the tool keeps working if you go offline after the page has loaded. We only count that a download happened, never the file or its name.

Binary STL files up to 600 MB (about 12 million triangles), ASCII STL up to 600 MB, and OBJ or 3MF up to 150 MB. The file is read in chunks, so it isn’t held in memory twice. As a guide, on a desktop computer: a 100 MB STL (2 million triangles) opens in about 1–2 seconds and is repaired, merged and thickness-checked in about 8 seconds; a 400 MB STL (8 million triangles) is repaired and checked in about 20 seconds using over 1 GB of memory. The geometry engine that merges parts and verifies the result handles up to 4 million triangles – above that, turn on Simplify and the model is reduced first, then merged (8 million → 1 million triangles takes over a minute). Phones and tablets have far less memory: expect a few hundred thousand to about a million triangles there. If the browser runs out of memory you get a message, not a crash – export with less detail, or in parts.

That is what the merge option is for. Architectural exports are usually hundreds of separate boxes and surfaces – walls, slabs, columns, mullions – that overlap each other, with some faces flipped, some missing and some doubled. The tool welds and closes every part it can, turns inside-out parts the right way, removes coincident and internal faces, then unions all the closed parts into one solid with the manifold geometry engine, so the slicer sees one clean shell instead of a pile of overlapping ones. Surfaces with no thickness at all (a glass pane drawn as a single face, terrain as an open sheet) can’t become solid: they are listed, and you choose whether to keep them as separate surfaces or drop them.

Pick the unit it was modelled in (the tool guesses metres, centimetres, millimetres or inches from the size) and a scale: 1:100 is typical for a single building on a normal printer bed, 1:200 or 1:500 for a block or a site, 1:50 for details. “Fit the longest side” scales it to a size you type. The final printed size is always shown in millimetres, and the filename says which scale you chose.

At 1:100 a 120 mm brick wall becomes 1.2 mm, a 50 mm mullion half a millimetre and a glass balustrade almost nothing. The thickness check casts rays through a sample of the faces and marks in red every place that is thinner than the limit (1.2 mm by default – about three lines of a 0.4 mm nozzle). It only reports: thicken those parts in the CAD model, print at a bigger scale, or accept that they won’t print.

It reduces the number of triangles (to a percentage or an exact count) by collapsing edges where that changes the shape least, and keeps sharp edges and corners. Dense meshes – terrain, scanned facades, smooth curves exported finely – often have far more triangles than a printer can show; halving them makes a smaller file that slices faster. Very strong reductions round off small details, so compare Before / After.

Yes, as an experimental option: the solid is turned into a fine grid, the inside is offset by the wall thickness you choose and carved out, and the result is checked by the geometry engine. It needs a merged, closed solid, uses a grid of at most 150 cells across, so walls come out within about one cell of the chosen thickness, and it does not cut drain holes – add those yourself for resin printing. For FDM printing, your slicer’s infill setting is usually the better way to save material.

Architectural exports draw many walls, floors and windows as single faces with no thickness. A repair can close holes, but a face with no thickness can’t become solid – so “Drop open parts” leaves a skeleton. Tick “Thicken surfaces that have no thickness”: every flat surface becomes a slab of the wall thickness you set (1.2 mm prints well with a 0.4 mm nozzle), and the building stays whole. Leave Merge off for big models: the slabs overlap and every slicer fills overlapping parts; merging them into one solid is possible but slow.

It doesn’t rebuild missing detail – a filled hole is closed with a flat patch – and on its own it can’t give thickness to surfaces that have none – “Thicken surfaces” turns those into slabs of the wall thickness you choose. Badly torn 3D scans, or parts that are open on many sides, may not close; the report says exactly what is left. Self-intersections between closed parts are resolved by the merge. A part that crosses itself is found and shown in amber (models up to 1.5 million triangles); “Rebuild parts whose surface crosses itself” rebuilds just that part as one clean solid on a grid sized to it, rounding off detail smaller than one grid cell; a part too thin for its grid is left as it is and named in the report. For heavy mesh surgery, use Blender’s 3D-Print Toolbox or Meshmixer.

Binary and ASCII STL, OBJ (polygons are split into triangles) and 3MF, including 3MF files from Bambu Studio, PrusaSlicer and Windows. You download a binary STL in millimetres that every slicer opens – Bambu Studio, PrusaSlicer, OrcaSlicer, Cura.

Only if you turned on “Split for printing”. Otherwise the repaired STL keeps everything in one file. “Split into separate STLs” is different again: it saves each loose part that is already separate as its own file, for example to print parts in different colours.

Yes. Turn on “Split for printing” and list your printers – bed size and how many of each, for example three Ender 3 (220 × 220 × 250) and one K1 Max (300 × 300 × 300). The tool finds the grid of straight cuts with the fewest pieces that still keeps every printer busy, turns pieces 90° (or onto their side, if you allow it) so they fit, moves each cut within a search window to a thinner section so there is less to glue, and shares the pieces out by estimated print time so the printers finish at about the same time. You can also set the number of cuts yourself and drag each cut plane. Every piece is checked watertight, and each one is checked to fit the bed of the printer it is assigned to, with the clearance you set.

Every cut face gets two to four alignment joints, laid out unevenly so the pieces only go together one way. Choose pin holes on both sides with separately printed pins (4 × 16 mm by default; spares included), holes for 1.75 mm filament offcuts, a peg on one piece with a socket in the other, or pockets for 6 × 3 mm, 8 × 3 mm or other magnets. The tolerance (0.2 mm by default) is added to every hole. Joints are kept inside the solid part of the cut face, with a wall all round, so none breaks through to the outside. On thin architectural walls, where a 4 mm pin has no room, that face falls back to 1.75 mm filament pins. A face too small for any joint is listed so you can glue it flat.

Each piece gets an ID – row letter from the front, column number from the left and a level number when stacked, such as A1, B2 or A1.2. It is engraved (or raised) on the bottom that sits on the bed, or on a cut face, never on a visible outside face. Next to the joints, both faces carry a matching mark such as A1|A2. The files in the ZIP are named the same way, with the printer: A1_Ender-3-2.stl. You can switch labels off.

One STL per piece, named with its ID and printer, a pins STL when you use printed pins, a parts list (CSV) and the assembly sheet. The sheet is a printable page with a top view of every level and a front view, coloured by printer, the cut positions, which printer prints which piece, the joint type and how many pins you need, and the size, volume and a rough print time of each piece. Print it, or save it as a PDF from the browser.