Check "Known issues and workarounds.txt" in the download first - it is written plainly and with measurements, and covers more than this page can. If that does not answer it, contact Dev.Luv Studios through this product page or at [email protected]. Telling us your Blender version, which of the two downloads you installed, and what the L/R Balance panel says will get you a useful answer fastest.

Mostly you should not need to; the defaults are the tested path. Two worth knowing: leave "Cache Symmetry Mapping" ON for Mirror All - switching it off is dramatically slower AND loses weights, so it is not a speed-versus-accuracy trade. And do not enable "Auto-Clean Micro-Noise" together with "Invert Weights"; the panel shows a red warning when you do. The shipped known-issues document lists every combination worth avoiding, with measurements.

On hair-card, lash and brow meshes this is a real trade-off rather than a bug, and nothing has been destroyed - those weights are never written, and your source group is untouched. Depth keeps the asymmetric fallback off, so normal-alignment rejection takes full effect; on flat two-sided cards roughly 40% of vertices have normals facing away from their mirror. If you want the count back, untick "Verify Normal Alignment" and mirror again, leaving Depth on. If the weights look right and only the count looks low, leave both on - on a card mesh a lower count is the expected result. On ordinary body meshes none of this appears. One caution, because this answer and the one before it point opposite ways about "Verify Normal Alignment" and both are right. On hair cards the rejection is spurious - a flat two-sided ribbon really does have its back facing away from its mirror - so you untick it. On lips and eyelids that same rejection is the entire point, so you tick it. The filter has not changed; what differs is whether the surfaces facing apart are one surface or two.

Turn it on for garments, sculpted and asymmetric meshes; leave it off for stacked shells. It ships off. Its contract is that no weights disappear: where the two halves are not built vertex-for-vertex, the only available counterpart often faces the other way, and refusing those grabs throws away correct weights - on the garment fixture, about 500 per group, silently. It also overrides "Verify Normal Alignment": while the fallback is on, that filter has no effect, so it only protects anything with the fallback off. Keep it OFF, and turn Verify Normal Alignment ON, when you are mirroring lips, eyelids or anything with two surfaces close together - that pair is the anti-leak setting. "Confine to Mesh Island (Depth)" keeps the fallback off, switching it off if you had it on, which is what the next question is about. How much it recovers, measured on the asymmetric fixture with the target emptied first: source 678, mirrored side 378 with the fallback off and 671 with it on - 55.8% of parity to 99.0%. Worth knowing WHY: with Verify Normal Alignment switched off, the plain matcher reaches the same 671 on its own. The fallback is not an approximation, it is what undoes that filter's over-rejection on two-sided geometry, where the back of one card faces away from the front of its mirror. It does not invent symmetry, and it does not quite reach parity: seven vertices of 678 on that fixture stay unwritten. We checked whether they simply have no counterpart and could not show that they do not, so the honest answer is that we do not yet know why those seven.

That is "Restrict to Counterpart Vertices (Mask)" showing you its work. With that option on, Mirror Active Group writes only onto the true counterparts of the source group's weighted vertices, and leaves exactly those vertices selected with the paint mask on - so you can see what changed and paint straight onto it. It is not a leftover and it will not affect your next mirror. The add-on stamps the selection it made and ignores its own stamp, so a mask one operator leaves behind cannot quietly shrink a later one. Change the selection yourself and the stamp goes stale, at which point the mask counts as deliberate and is obeyed again - which is what you want when you are masking on purpose. Three buttons work this way: Restrict to Counterpart Vertices, Select Unweighted/Missing Target Vertices and Select Unpairable Vertices. The option itself applies to Mirror Active Group only; Mirror All and Quick Fix work on the whole mesh by design.

The first draw on a new object runs the full analysis; after that it is effectively instant until something changes. On an ordinary body this is under a second, and on the most awkward meshes - dense, many-grouped, overlapping shells - it can be a few seconds. A big character that is not built symmetrically takes longer: measured on a 70,000-vertex Daz character with 187 groups, about 20 seconds the first time and about 11 after a Mirror All; every redraw after that was instant.

Clean Unmatched is off by default, so a side you mirror onto keeps any weights of its own that have no counterpart on the side you mirror from. Mirror All says how many pairs did. Turn Clean Unmatched on if those weights should go.

Its time grows in step with the vertex count. Measured on a 14,000-vertex Rigify body with 167 groups, split into ever finer grids: about 3.5 seconds, 13 at 56,000 vertices, 52 at 225,000 and about five minutes at 1.1 million. From 10,000 vertices Mirror All shows its progress in the status bar, and Esc cancels it with every weight put back. If you can, mirror before you subdivide: the weights carry through subdivision. With Clean Unmatched on, every pair came out an exact mirror up to 225,000 vertices. At 1.1 million, 22 of the 74 pairs - in the face and three fingertips - had some weights a few vertices from their twin; Mirror Active Group was exact there.

It is working, not crashed. Those meshes are thousands of small overlapping pieces and the matching work is genuinely large - on a dense eyebrow mesh a repair can take several minutes. SymmetryFlow warns you before it starts one of these, so you can save first or cancel.

The balance box caches its analysis and recomputes when an add-on operator runs, not when you paint. Run any SymmetryFlow operator - Run Diagnostics is the cheapest - and it will refresh.

On face regions, check the vertex COUNT before assuming something went wrong. Character faces are frequently not vertex-aligned across the mirror plane, so a rebuilt side can carry the same weights spread over a few more vertices than the original had - an upper eyelid group going from 35 to 43, for example. The weights are correct; the tally differs. Judge those regions by the weights and the viewport, not the count. If you need the counts to agree exactly, fix the mesh first: Merge by Distance, then Mesh > Symmetry > Snap to Symmetry, then mirror again.

This one surprises people, and it is deliberate. When the operator finishes it makes the COUNTERPART the active group, so you are looking at the side it just wrote. Clicking again therefore mirrors back the other way. On an already-symmetric pair that is harmless, but if you edited the original source in between without re-selecting it, those edits are overwritten. Select the group you want to mirror FROM immediately before each click. When you are weight painting this happens naturally, because selecting a group in order to paint it also makes it active.

Yes. One Ctrl+Z reverts a mirror or a Quick Fix completely, including the multi-stage repairs, and redo works too.

Select your mesh, enter Weight Paint mode, open the SymmetryFlow tab, make the group you want to mirror the active one, and click "Mirror Active Group". For a whole rig at once, "Mirror All L/R Groups". Then read the L/R Balance panel: anything reading Mismatched has a Quick Fix button.

Two usual causes. Either the add-on is installed but not ticked in the add-on list, or you took the download for the other Blender generation - in which case Blender either refuses the file or the add-on appears but will not switch on. Nothing is broken; install the other one. The panel itself is at 3D Viewport > press N > "SymmetryFlow" tab, in Weight Paint or Object mode.

No. Blender installs the add-on from the zip, and will not find it if you extract it yourself. Edit > Preferences > Add-ons > Install from Disk, and point it at the zip. On Blender 3.6-4.1 the same entry is called "Install".

Yes. SymmetryFlow is GPL-2.0-or-later and the source ships in the download. The full manual, the known-issues document and the complete changelog are in the download as plain text.

No, Where one side has no matching vertex on the other, there is nowhere to put the weight exactly. What SymmetryFlow does instead is tell you which groups those are, by name, rather than quietly doing something wrong. One option, "Robust Asymmetric Fallback", recovers part of the gap - the counterparts a strict check refuses. On the add-on's own asymmetric fixture it takes the mirrored side from 56% of the source's weighted vertices to 99% - 378 of 678, then 671 of 678. How close it gets on yours depends on the geometry. The last seven of 678 it does not reach, and we have not established why on that mesh. "Using it" below says when to leave the option on.

Yes. That is exactly where a coordinate mirror fails, because the two sides are in different places in space. SymmetryFlow's topological engine follows the mesh's own edge flow instead, and Auto-Detect will normally choose it for you.

That is what it is built for. It handles the usual side-naming conventions out of the box: _L/_R, .L/.R, -L/-R, Left/Right, Lft/Rgt and the matching prefixes, plus Daz Genesis 3 / 8 (lShldrBend / rShldrBend), 3ds Max Biped (Bip01 L / Bip01 R) and MMD's Japanese names. If your rig names sides some other way, rename them or set a custom match pattern in the panel. Until you do, Mirror All leaves alone any group it cannot pair whose weight is all on one side, and names it, rather than copying it onto the other side.

No, and please don't install it there. Blender 5.0.0 and 5.0.1 contain a bug in Blender itself: any add-on that highlights vertices and then leaves Weight Paint mode closes Blender instantly, with no error, losing unsaved work. We confirmed this is Blender and not us by reproducing the crash with SymmetryFlow completely uninstalled. Blender fixed it in 5.1. Use 5.1 or newer, or 4.2 LTS, 4.4 or 4.5.

3.6 LTS, 4.0.0, 4.2.9 LTS, 4.4.0, 4.4.3, 4.5.0 and 5.1.2, with 5.1.2 as the primary development target. Builds in between - 4.1, 4.3 - install and run on the download for their side of the 4.2 split; they have not had the test suite run on them.