Retopology tools are having a moment right now, and it says something that I nearly wrote about the boring one. Today’s feeds flagged Quad Maker, a manual quad-drawing addon for Blender modelled on Maya’s Quad Draw — click, drag, place a quad, repeat. That’s the honest, unglamorous version of the job, and it’s had an update this week. But digging around it turned up something a lot stranger sitting one tab over: Quadify Ultra, a Blender 5.0 addon that doesn’t retopologize your mesh so much as decide how it should be retopologized, before it touches a single vertex.
What it actually does
Quadify already existed as a straightforward auto-retopo tool. Ultra adds a routing layer on top: select a mesh, hit a hotkey, and it analyses 18 geometric features of that specific object — curvature, edge density, hard-surface breaks, that kind of thing — then picks which of five different algorithms to run: a general Smart mode, a Hard Edge Preserving mode, Dissolve + Requadify, QuadriFlow + Transfer, or Voxel Remesh + Transfer. It shows you a confidence score before committing, and it claims to preserve your existing UV seams, material boundaries and sharp edges through whichever path it picks — that’s the part separate from most one-click remeshers, which tend to nuke all three.
The developer describes it as a KNN-based routing engine rather than a single trained model, and there’s a Community ML Model option: opted-in users can contribute anonymous telemetry on which algorithm worked for which mesh type, and the routing supposedly gets sharper for everyone as more people use it. Posted production numbers — hard-surface vehicle panels landing at 88–97% quads in under a second — are the developer’s own benchmarks from the release thread, not an independent test, so read them as a claim rather than a verdict. It’s a $200 one-time purchase, Blender 5.0 and up.
The part worth sitting with
Automating the retopology pass isn’t new — QuadriFlow, Instant Meshes and Blender’s own Voxel Remesher have been doing versions of that for years, with predictably mixed results depending on the mesh. What’s new here is automating the decision about which of those approaches to use, which is a different kind of task entirely. Picking the right algorithm for a given mesh is exactly the bit of retopology that actually requires judgment: a hard-surface gun barrel and an organic creature torso do not want the same treatment, and knowing that on sight is most of what separates a technical artist from someone who just mashes the remesh button and hopes. Quadify Ultra is proposing to fold that judgment call into the tool.
I don’t think that replaces the skill — a router that’s wrong 12% of the time on its own published numbers still needs someone who can look at the output and know it’s wrong, which means it needs someone who already understands why each algorithm exists. But it does change what the entry-level version of that judgment looks like. The thing junior artists used to build by retopologizing a hundred different mesh types by hand, slowly learning which method suits which shape, is exactly the thing this tool is trying to shortcut. That’s a genuine loss even if the tool works well, the same way spellcheck is a genuine loss for anyone who never has to learn to spell.
The telemetry angle
The Community ML Model is opt-in and anonymous, per the listing, and I have no reason to doubt that as stated. But it’s worth being clear-eyed about what it actually is: every participating studio or freelancer is quietly training a shared model on the geometric fingerprint of whatever they’re modelling, in exchange for slightly better routing decisions. For an indie prop pack that’s a non-issue. For anyone under an NDA on an unannounced game or film asset, “anonymous” doesn’t automatically mean “fine to opt into” — check what your contract actually says about sending derived data off-machine before you tick that box, not after.
Where I’d actually use this
Background and mid-ground hard-surface work is the obvious fit — the vehicle-panel numbers in the release thread are believable for exactly that reason, since hard-surface geometry with clean curvature breaks is the easiest case for any router to get right. I’d be far more skeptical running it unsupervised on hero organic topology, faces especially, where the “right” edge flow is about deformation and animation needs the router has no way to know about. Worth trying on the parts of a scene nobody’s going to be looking closely at; worth checking by hand on the parts they are.
This is based on the developer’s release notes and forum threads (Blender Artists, Polycount) covering Quadify Ultra’s launch — I haven’t run the addon myself, so treat the accuracy numbers above as the maker’s claim rather than something I’ve verified in production.