gittrahan/support-fins

Tip a part on edge, and Support Fins adds the breakaway support fins that make that orientation printable — baked right into the STL.

Live at printfins.com. Runs entirely in your browser: nothing uploads, nothing installs, no account.

Printing a part flat is usually the weakest way to print it. Lying or diagonal layer orientation tests up to ~3× stronger than standing up. Most people print flat anyway, because the strong orientation needs supports, and slicer supports scar the surface, waste plastic, and take longer to pick off than the part took to design.

Designed-in fins fix that, and they beat slicer supports in one way a slicer can't touch: the support lives in the STL. Upload it anywhere — any printer, any filament, any slicer — and it still comes out right. A slicer only ever outputs gcode for one machine.

No slicer generates these. OrcaSlicer's whole style list is Grid / Snug / Organic / Tree Slim / Strong / Hybrid, and none of them modify the mesh or bond to the part on purpose. The technique isn't new (Slant3D has evangelized designed-in supports for years), but until now you had to CAD it by hand every time.

1. Import an STL, 3MF or STEP.
2. Rotate it. You're in control — Support Fins suggests, it never decides for you.
3. It shows you live: overhang count, how many can take a real fin, height, bed contact. Point at the load direction, answer one question — does it pull apart, or does it lever? — and it scores orientations for strength too.
4. Export. Fins and a bed pad come baked into the STL (or 3MF).

Why you pick the rotation, not the software: "stronger" means nothing without a load direction, and the geometry doesn't contain one. Turn a solver fully loose and it'll hand you a part 155 mm tall balanced on a needle with two sail-sized fins — technically optimal, completely unprintable. (Ours did exactly that.) So the human makes the one call the software can't, and the software does the rest.

The web app is live and does the full loop: load, rotate, score, fin, export. The geometry engine is validated against third-party STLs and pinned by an offline test suite.

Working: overhang detection, bed-reachability, contoured breakaway fin walls, orientation + load-direction scoring, the combined fin (wall + tines that fuse into the part — the whole point; see docs/FIN-SPEC.md), optional sway braces that tie tall parts' sides on all the way up (auto, or click an upright side in Draw), STL, 3MF and STEP import, STL and 3MF export.

Still open: scale-aware fin profiles, and the bed pad on tilted exports.

Set Nozzle (0.2–2.0 mm) to your printer's nozzle and the support widths follow it; Wall lines (2 / 4 / 6 / 8) makes walls, wedges and sway braces thicker for tall parts. The contact tip and tines stay one line wide either way. See the nozzle guide for the sizes and what to check in the slicer.

Full brace depth is an optional checkbox under Sway. It starts off, keeping the usual 60 mm depth limit. On, Brace depth uses its selected 5–50% of the actual brace height beyond that limit: a 999 mm brace at 15% reaches about 150 mm. The wider foot needs more plate space and material; model/support collisions can still shorten or refuse a brace. Check the build-volume preview and sliced first layers. Thickness, contact tines and grip rules are unchanged; deeper metre-high braces have not been physically print-tested. Undo/redo and 3MF sessions retain the checkbox. Direct callers opt in with sway: { on: true, uncappedDepth: true }.

Tall, slender parts have a problem the fins were never built for: nothing overhangs, but as the part grows, the nozzle's drag and each layer shrinking as it cools push the top around. The part drifts, sags or wobbles, and every movement shows up as a layer line. Sway braces stop that by tying the part's upright sides to a stiff support all the way up.

What a sway brace is: a vertical rib standing edge-on to an upright side (its stiff direction). It's deep at the bed and tapers to a 4 mm flat top, gets thicker as it gets taller, and sits on a thin foot on the plate. One-layer horizontal tines, spaced evenly up the full height, tie it to the part. Like every support here, it stands off by the breakaway gap and snaps off; only the tines touch the part.

Using it: tick Sway braces (tall parts) in the options panel. It's off by default.

• Auto braces the tallest sides for you: up to four faces facing different ways, so both axes are held, with each rib placed where its face reaches highest.
• Draw: one click on an upright side stands a brace there. Click a support you placed to select it (amber), then press Delete or Remove selected; Undo brings it back.
• Three settings appear while it's on: Brace grip from (height the tines start; 0 = the whole height), Brace tine spacing (default 6 mm) and Brace depth (% of height at the bed; default 15%).
• Braces keep at least 1 mm of air between them. One that would run into another, for example straight across a narrow channel, is refused with a reason rather than fused into a bar that won't break away.

Why this shape, not the old Brace fin: the Brace fin lies flat against the face, so it bends the easy way exactly when the part leans into it, and its tines bunch at the base and spread out going up, leaving the top of a tall part, where the sway is, nearly untied. Every number and the reasoning behind it is in docs/FIN-SPEC.md ("Sway braces"); the code is web/sway.js, and tests/sway.test.js pins its behaviour.

Status: printed. Developed on a 249 mm fence-post cap, where Auto places 4 braces (about 20 g of support) and hand-placed braces follow its gable up to 225 mm. Two test prints (2026-09-22) both came out clean, so the defaults below — depth, thickness and tine spacing — are the printed ones, not estimates.

Opening the 3MF, Bambu Studio ("invalid config, load geometry data only") and PrusaSlicer ("does not contain PrusaSlicer configuration. Only geometry was loaded.") show a notice and import just the mesh. This is expected and harmless — every slicer shows it for any geometry-only 3MF (Fusion 360, FreeCAD, even the 3MF Consortium's own reference files). The part imports correctly oriented and sized; the fins come in as intended. Just slice with supports off. (OrcaSlicer opens it without a notice.)

The export ships pure geometry with no slicer profile embedded on purpose: baking in a profile would silence the notice but replace whoever-opens-it's printer/filament/ print settings with ours on load, and it would have to be re-authored per slicer and per slicer version — a worse trade than a one-time, benign notice on a file whose geometry is already right. See web/threemf.js for the writer.

The web app is vanilla ES modules — no build step. Serve it with the included dev server (it disables caching so edits actually show up on reload):

python3 dev-server.py                      # http://localhost:8731/
python3 dev-server.py 8080                 # custom port
python3 dev-server.py --host 0.0.0.0       # reach from other devices on the LAN

By default the server binds to 127.0.0.1 (localhost only). Pass --host 0.0.0.0 to expose it to the local network — handy on a headless box like a Raspberry Pi behind a firewall; the script prints the LAN address to open. The port is an optional positional argument and --help lists every option.

Or run the same web/ directory in Docker — nginx on the host's 8731, so the URL is identical to the dev server:

docker compose up --build        # http://localhost:8731/

There's no build step and no backend, so the image is just nginx:stable-alpine serving static files with cache headers that match the dev server. See docker-compose.yml, Dockerfile, and nginx.conf.

The Python prototype is the proof of concept the engine was ported from — plain mesh math, no CAD kernel:

pip install trimesh numpy manifold3d
python3 prototype/spike_overhangs.py yourpart.stl      # what needs support
python3 prototype/spike_fins.py yourpart.stl out.stl   # add fins
python3 prototype/spike_orient.py yourpart.stl         # rank orientations
python3 prototype/spike_arrow.py yourpart.stl 0,0,-1   # load-direction scoring

Tests (Deno for the JS engine):

deno test --allow-read tests/

plugins/prusa/ is a native PrusaSlicer 3.0 plugin, Support Fins → Add a Fin. The 3.0 plugin sandbox can't read a loaded mesh's triangles, so it can't do the automatic tool. It drops one angled-print support fin instead: a thin triangle whose slope you set 0.2 mm under a tilted part's underside, with one-layer tines along it (Slope Angle, Fin Height, Tine Spacing). You place it by hand; size it with Fin Height rather than the slicer's scale tool, which would stretch the one-layer tines. Confirmed working in PrusaSlicer 3.0 alpha11 (2026-10-02). For fins shaped to the part automatically, use the browser app. See plugins/prusa/README.md.

• Overhangs sitting over the part rather than the plate aren't handled — fins attach to the bed only.
• Features shorter than roughly a 4 mm wall height are too short for a real fin.
• It won't pick your orientation for you. On purpose.

web/         the browser app (live at printfins.com)
plugins/     slicer/CAD integrations (PrusaSlicer, OrcaSlicer, Onshape, Autodesk Fusion), the
             command line, and an MCP server for AI clients
prototype/   Python/trimesh proof of concept the engine was ported from
docs/        FIN-SPEC.md — the verified fin geometry, with sources
tests/       offline geometry regression suite

Fins are separate closed solids appended to the mesh; the slicer unions them. The whole engine is plain mesh math with no boolean kernel, because it has to run in the browser.

The fin technique is Slant3D's — they've evangelized designed-in supports for years. Support Fins just automates it. docs/FIN-SPEC.md cites their numbers directly.

MIT. The license covers this tool, not what you make with it — STLs you run through Support Fins are entirely yours, and the output carries no license obligation.

STEP import uses occt-import-js (Open CASCADE compiled to WebAssembly), vendored unmodified under web/vendor/occt-import-js-0.0.23/ with its LGPL-2.1 license files. The browser only downloads it when you go to import a file.

Pointer picks on dense parts use three-mesh-bvh, vendored unmodified under web/vendor/three-mesh-bvh-0.9.15/ with its MIT license.

Free and open source. If it ever saves you a print, you can buy me a coffee on Ko-fi ☕.