Will Sub-D provide new and better ways to fillet?

I’d just like a quick opinion on whether Sub-D is likely to lead to quick routine ways to handle advanced fillets, without manual fixes. Examples I’m thinking of include the ‘tangent cylinders’ problem –
and the ‘overlapping boxes’ problem –

If modeled with well thought topology, SubD does not require any fillets.




Thanks for your video, @Inju. It stopped short of any actual filleting, but I do see that Sub-D seems to be an entirely parallel system to the use of nurbs surfaces and solids.

I probably should not have added that screenshot of mine, it seems to be distracting from my core question, so I am going to modify my question now by deleting the screenshot.

‘tangent cylinders’

‘overlapping boxes’


@Inju – At first sight, that seems to be a pretty good approach to the ‘overlapping boxes’ example, avoiding the manual patching required when working in nurbs. I’m assuming that there would be a way to numerically control the amount of fillet, e.g. if you wanted a uniform 0.05 inch fillet.

I’m less convinced about the tangent cylinders, it looks as though the upper cylinder may have become distorted above the fillet area, especially at the ‘back’ where the two cylinders are tangent.

@martinsiegrist – I’m not sure I could see my end product clearly enough at inception to end up with built-in rounded edges and intersections that don’t need filleting as a final step. I have enough trouble manipulating sharp-edged primitives and extrusions then Boolean those into the forms I need – predominantly mechanical parts that need to be dimensioned into an inter-related assembly typically of 20 to 100 parts. Kind of 3D sketching with solids. The fillets come at the end as a cosmetic touch after all the mechanical logic is settled.

In that case, you would probably stick to NURBS and fillets.

Sorry, but SubD is not for accurate modeling, if you need accuracy and tolerances, then you need NURBS.


Oh, wow – that probably dashes my hopes that Sub-D will provide a better way to perform tricky fillets on the digital prototypes of cast or printed metal parts.

I don’t fully agree on this. Especially when a part is cast or printed, SubD is ideal for all sorts of things.

@Ian, it requires a different approach than conventional NURBS modeling.

1 Like

I agree with you, NURBS for control, accurate models, a SubD for topology, for clean surfaces.

and so on.


Keep in mind that in Rhino “Fillet” means very specifically a circular (fixed radius) curve or surface profile.

It does not mean generally smoothed over.


Great point by John here. Fillets are not more accurate than SubD blends. They just happen to always have the same exact curvature/radius value along its surface.

Also keep in mind that fillets are fine for internal parts and components, yet, IMO, never acceptable for finished designed surfaces, unless you have very low aesthetic and continuity standards. Or are things so small and/or unimportant that no one will see they look like a shitty fillet.

Re: accuracy. If the topology of a model is 100% repeatable and measurable it is mathematically accurate. If a SubD model can achieve that it is accurate too. If the complexity of measuring its curvature is way more than the complexity of grasping a single value of a radial/conical surface, it does not mean is not accurate, it means that its accuracy is a lot more complex than your understanding of topology.


1 Like

Precision or tolerance is not in itself accuracy, it is a technique and tools to achieve a certain accuracy in the task at hand, laboriousness, and labor to perform this task.
If you do not feel sorry for the time spent, then on SubD you can make accurate models.

I suppose the point is that fillets (fixed radius surface profile) are easily documentable, particularly on drawings.

A secondary point might be that uniform radii can look sleek on traditional industrial objects, whereas haphazard curves might not look the part on objects that don’t have an organic aesthetic.

Edit: considering also the Autodesk explainer that @inju pasted above, maybe a third point is that if Sub-D fillet-like surfaces cannot be sparsely specified (e.g. specified as simply as R=5.0 in the case of a nurbs fillet) then they won’t work well in a parametric application.

I’m beginning to see that Sub-D surfaces, although maybe avoiding the time-consuming horrors of trying to fillet a complex nurbs object that will be cast or laser sintered, has its own time-consuming problems if you attempt ‘constant radius’ uniformity. Also as I see in the ‘tangent cylinders’ example above, there seems to be a problem in avoiding altering the geometry of the cylinder further away from the ‘quasi-fillet’. Surely all these problems can be solved, but the time consumed is the issue.

Very true a few decades ago. Today, at least in industrial design, drawings are mostly about general dimensions, and contractually all tooling releases are based on “As documented in 3D file”

This 2011, probably designed and modeled 5 years earlier, engine disagrees. Only a handful of parts here have fillets:


Sadly, I’m stuck in bygone days because I’m working with historical designs and objects, cultural heritage and museum work. So drawings make sense still, due to their archivability. And the objects conceptually assumed fillets, though granted in practice it came down to the patternmaker’s art. I had hoped Sub_D could step in as an adjunct to the basic nurbs geometry where lineal fillets meet one another in awkward ways, or terminate in a blend (e.g. in the tangent cylinder problem).

If I were to model this engine in Rhino, I’d consider most of those parts have rounded-off edges that could be interpreted as small-radius fillets. If I left the edges ‘sharp’ on the basic geometry I built up in nurbs, it wouldn’t look quite like the engine in the illustration. I’d still somehow have to round off the edges and intersections on those molded and cast parts. They’re probably using Catia or similar to do theirs. I remember years ago reading posts here where people compared the superior filleting ability of those high-end packages to Rhino.

Consumer awareness?

More Maximum fillet value… :thinking:

FILLET LIKELY~ Sorry G :slightly_smiling_face:

I don’t know if you are being sarcastic, but do you know there isn’t a single fillet in that phone right?



@John_Brock No longer always true.

In V7 a G2, non-circular arc blend option was added to FilletSrf. I had a discussion with @pascal about using “fillet” in Rhino for a non-circular arc blend. He said we can yell at him. Shrinking fillet surfaces

Back to FilletSrf in V7: The description of the command in Help has not been updated to include the option but it is shown in the Command line options. https://docs.mcneel.com/rhino/7/help/en-us/index.htm#commands/filletsrf.htm



Creates a fillet surface with arc sections and tangent (G1) to the input surfaces.


Creates a blend surface connecting to the input surfaces with curvature (G2) continuity.