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

A Blend surface or curve can be G2.
A Fillet surface or curve can only be G1 at best.

Sadly? That sounds like really cool work!

It’d be great to see a typical project requirement: what reference you start from and what you need to produce. I’d be happy to provide feedback. Here or privately.

Last year we did a project to digitize some very old classic hand tools, starting from wax masters that where hand sculpted by hand. They had first tried a reverse engineering software that would try to place primitives and fillets everywhere and of course that would not work. Every surface was a beautiful hand sculpted and polished G2 form. We needed to retopologize if all in SubDs.

Each challenge needs different approaches and general rules rarely apply in high craft work.


1 Like

Hi G~

Yes, I know Iphone curvature.
I thought came up your Reply that case study in Iphone curvature.

Thank you G

1 Like

True, I don’t feel too sad working on these projects – to me they are really interesting. But then, I do look across to the people working in manufacturing design and think that would be just amazing to be involved in.

Anyway, here’s a couple of pics from what I’m currently working on – a 50hp ‘Frisco Standard’ marine engine from 1916, with an open crankcase and exposed valve gear. Pretty steampunk really, and totally not workplace safety approved. The original engine as you can see is pretty decrepit, missing a lot of parts. It’s in a folk museum in New Zealand. There’s only one other 3-cyl Frisco in the world, in a private collection in Oregon, but this is the bigger one. At one time, there were hundreds across the west coast of the US and across the Pacific, into Australia and NZ. Now just the two remain.

The second pic is the 3D model just of the exhaust manifold (no internal detail yet, I’ve moved on to roughing out other parts of the engine). The documentation for this engine is to support a restoration that may only extend as far as replacement parts in 3D-printed nylon for visual effect, or possibly will extend to proper parts in metal. The model can provide the STL files for either, as well as 3D views and 2D drawings that archive the assembly, e.g. for anyone working on the engine. I’ll also use the 3D views downstream in two books I’m slowly working up.

You mentioned automated drawings from 3D-scanning as being hopeless. I’d agree, these antiques are so gnarly and irregular that a point cloud would only be a reference object. I use measurements, notebook sketches and plenty of photos. I often need to line up photos over the cad model using a transparent image viewer, just to get missing dimensions (e.g. fillets!) more or less correct. A lot of the challenge in an engine is fitting everything by mechanical logic, e.g. gears of specific parameters, proper timing angles, and of course everything aligning. The end result is never exactly what my notebooks describe – there always seems to be something that has to be nudged a little to get it to ‘fit properly’. The result is an idealization, kind of a new manufacture, that gets to the underlying concept of the machine as much as its surface appearance.

Third pic here is merely for curiosity value - it’s a mock up from some weeks ago of the engine, looks good at first sight, but is very incomplete.

So that’s where the issue of filleting fits in. So far I mostly don’t fillet things because there are often many intersecting surfaces that make automated filleting impossible. So the most time-effective solution is to not go there unless its easily done, and just accept the images look ‘pretty good’, certainly to casual viewers. Any reproduction parts obviously have to be worked on much more. But, if better auto-filleting – or an alternative approach – ever turns up, I’ll be jumping on it.

Sorry, raved on a bit here and might not have even answered your main inquiry. Maybe treat this post as a magazine segment.


@John_Brock That is no longer the case in Rhino for BlendSrf in V7.

Hi Ian,
Nice projects, the Frisco 3-cyl engine looks doable in rhino without sub d using the standard nurbs tools. Looks like most of it is surfacing and not filleting. But yes we all wish for better fillets and faster solutions. I almost think it would take more time in Subd, but I would try both at the same time since you can in V7 easily take your models or parts of them back and forth and use the best of both worlds.

Then the really fun part for this project would be to export it to your 3dpaint program like 3dcoat or substance and make all those wonderful rusted aged textures.


What is precision?
Not accurate, and not round! Just a surface, it’s a SubD property.

While you are certainly technically correct, I think very often in cad related discussions the term “fillet” is used in a more general manner including g1 and g2 fillets.
Also the terms “rounds” and “blends” are sometimes used for fillets (while “blends” in Rhino are specifically g2 fillets)

This is very debateble and also a bit condescending IMO.

Depends on so many variables: material, production methods, the effect the designer wants to achieve ecc…
Sometimes the edgy engineered look of good old fillets may be exactly what you are aiming for.
For an expert designer “fillets or blend?” is a question of style not of quality.

Expert designer… whatever that means

If styling is your only goal, then yes, style over quality

I’m an industrial designer and I follow other principles.

“Expert designer” was meant to be a bit provocative. It worked.

I didn’t say that style is my only goal, I explicitly mentioned “many variables”.
But style, semantics, expression call it what you want are important components of design.
You may have other goals, and might ignore these aspects of design.
That doesn’t make them go away.
At the end of the day even the most austere “form follows function” design philosophy remains a “style”.

When I said “a question of style not quality” I was answering to @gustojunk 's statement that “g2 is better quality than g1 fillet” wich I don’t agree with (obviously).
I meant to say that there is no inherent “better quality” to blends over fillets. They have different uses based on various variables one of them being different stylistic expression.


hi @Ian , this is really cool!

I’m going to share an R&D project we did on retopology. We wanted to test 3D scanning (the state of the art in 2020) and retopology based on SubD (and compared to filleted primitives replacement of scan data).

I’ll be back later today or in the next couple of days with some workflows that you might find useful based on what you shared. Stay tuned, I’ll make sure I mention you.




Hi @Ian,

based on what you describe there…

…and the pictures you posted, I think there are a few things you should consider.

For 3D printing and visualization you really do not need Nurbs at all. So any SubD/mesh workflow will work for you just fine. Fighting with fillets for anything that are not extract mechanical parts of your model seems like a losing battle.

A very effective workflow is creating all the organic shapes in SubD, and then adding mechanical details in Nurbs (things like cylinders, flat planes, etc.).

Also I think you might have found out that some of those engines, since they were build from pattern masters (probably build by hand in wood/wax/plaster/etc) by hand they are not as precise as what you would do today in a CAD to CNC/3D printing workflow. Except probably for round shapes turned in a lathe.

SubD retopology is an excellent way to build these macro forms, before you add details with booleans. Also 3D scanning is a very effective way to capture those castings original surfacing results, as compared to their theoretical intent in their drawings.

Here’s an example of a part that had no cad and how we scanned it an retopologized it. This was an experimental project at my company Fresco Design, so it’s not confidential or under NDA, so we can share it out (unlike most things we do).

  1. start with a physical sample (not different than your engines)

  1. 3D scan it, modern scanners can give you a really nice watertight model with a few passes. And all the stitching work is quite simple these days (if you have a beast PC of course)

  1. Retopologize by planning your poly layout, and your loop flow. This IMO is the most important part, and what probably dictates how easy/hard will be later on to add details as you need them. This is also what takes most practice and skill.

We do our retopo work in Modo. You can also use tools like the wonderful $100 Topogun, or Blender has some neat tools for this too. I read here today that @theoutside is planning to start bringing some retopo tools to Rhino too. Besdies proper retopo tools, I know the Rhino team loves to praise the QuadRemesh tool too, I do not find it of much used for these kind of controlled shaped like man-made mechanical objects like these. Even using its curve guides seems extremely stubborn IMO, but maybe you have better luck? If anything, it’s still a great tool to get you 40-50% there, especially if you are not fast at SubD modeling. In our case we prefer to layout the flow of polygons based on the form we have to tackle for retopology, or the form we have in our minds that we are designing while modeling.

  1. here’s a Symmetry comparison of input (3D scan) vs output (Retopo SubD)

  2. With tools such as blendshapes/morphmaps you can add topology based detail that you want to explore it’s gestural state. For example for these castings we wanted to be able to add more/less shutline lip, this is where all the imperfections of the casting go to, and then in a secondary process you remove them form the metal part. Similar than some of the parts I see in your photos. So if you were to cast these parts in medal, you will want to bring back to your model this geometric state that was removed by grinding.

Rhino does not have blend shapes yet, or ways to store control points positions AFAIK. I was hoping that Bongo 3 would support these. Blendshapes are extremely useful for design iteration and for compensation of manufacturing processions like warping or shrinking. But you can at least manage the two start and end states with duplicate models, a horrible hack I know. I’d never do it myself but I’m mentioning it because I know that many people prefer (or are limited) to use Rhino for everything, instead of using the right tool for each task.

Another reason for SubD: it’s extremely friendly for high end visualization work, where you have a lightweight model with sensible placement for seams to UV unwrap all textures (you are also better off leaving Rhino for that task at the moment), and then you can bake all the rendering detail into low poly models too for Web Visualizations and AR.

Here’s a rendering done in Rhino/ Octane after being UV unwrapped in Modo:

and an animation in Modo:

That kind of animation work can easily be done in Bongo too, but then you could only export still frames, not geometry to other platforms where you want that animation to go, like here in Sketchfab on any browser:

or here in AR right on yours space (look at that QR code with a modern phone and follow the link):

There’s one more think you slould consider to your workflow too, This upcoming gem from @DanielPiker:

I hope this helps, please let us know if you have any questions.

from my team, @IgorK and @jsantocono were involved in this project too.


@gustojunk Thanks for the description of your workflow and the reasons for it. This it the type of information which adds fundamental value to the forum.

I create 3D models of boats, starting from either a mesh from photogrammetry using Metashape, or from traditional 2D drawings. Currently Rhino is the only software I use and I work only in NURBS. Since the boat hulls are built by springing planks around a set of frames/molds NURBS can provide a good representation of the shape. I retopologize (new word to me) using several Rhino tools including Patch with a perviously generated starting surface (never Patch directly by itself).

An essential element for creating a good model is to think about how the shape was created. Sometimes I’ll look for photos of the boat or a similar boat under construction or talk to someone familar with how similar boats were built. The knowledge of how the shape was created frequently guides decisions on the topology of the NURBS surface model which may be different than if I was designing a similar but new hull.

I need to learn to use SubD. Areas of boat hulls which are carved typically are difficult to model using NURBS. What I need to be able to do is to create geometry using SubD which intergrates with previously created NURBS geometry.

A final comment. The Lie-Nielsen spokeshave example is of particular interest to me for two reasons. I own the same spokeshave except it has an earlier version of the blade plate, and Lie-Neilsen is 15 minutes from us.


Hi G~

-Retopologize by planning your poly layout, and your loop flow.
This IMO is the most important part, and what probably dictates
how easy/hard will be later on to add details as you need them.
This is also what takes most practice and skill.-

I agree with your thought. and I solved to be subject this sentence.
I think that it is a part that requires a lot of consideration,
and it is a part to consider in depth by looking at
the model shape of the advanced rhino user.

Recently Kyle’s post official Rhino U-tube channel so nice and useful work-flow.

-In our case we prefer to layout the flow of polygons
based on the form we have to tackle for retopology,
or the form we have in our minds that we are designing while modeling.-

Philosophy? looks GREAT.
by looking at these results (Luxurious tool) and the work process.

Thank you for your comments
for detailed and much information.

Best Regrads

remember when using quad remesh… guide curves as edge loops, detect hard edges and careful attention to the polygon count can make a HUGE difference in how well the model meshes…

don’t forget, you can also project a subd to a mesh- (see here): https://www.youtube.com/watch?v=VSN3u2PgELk

Which allows you to select and delete bad quads that you may have made using quad remesh, then redraw the new quads you want in a flat plane, then project the new quads onto the mesh and stitch back into the existing SubD…

there is a lot of power there, even if it is not a readily discoverable workflow. I’m just starting to piece this stuff together myself. Since this is all new… we are all learning together.

Give it a go and see what you think. Ideas for improvement are always welcome.


Gustavo, your reply has a lot to unpack – I sense that there’s much more to it that I can appreciate.

At the moment, this seems to be the key point. All I need to know now, is what retoplogy means – people here are saying it a lot and I don’t understand to be fully honest with you. If it means to ‘trace’ a Sub-D surface coincident to a 3D scanner point cloud, I can get that. If it’s more than that, I don’t know yet.

But what I’m hearing is that there’s a radically different approach now, that turns everything on its head. I probably won’t be investing in a 3D-scanner at this point, because I doubt I’d get my money back – in the museum world, the fashion is rapidly shifting away from curation of industrial objects to exhibiting ‘narratives’ about such things as immigration, colonialism and people of color. These are the areas the upper level people in the institutions go for now. That’s just how it is, I’m not offering a judgment on that. But I do see that funding for what we are talking about, is drying up at an accelerating rate.

However I do get the sense that this new kind of workflow won’t be fundamentally dependent on using a 3D scanner. The comments made after your post suggest to me, especially confirmed by @theoutside comment:

that there’s a lot to learn, especially when I’m kitchenhand-level in a forum full of Michelin 5-star chefs. I’m not even quite sure what questions to ask. I think I need to see more workflows using these new tools and just get a feel for it all.

1 Like

IPhone 12 pro has a lidar scanner built in… I’m assuming we will be able to use this data as a basis for quadremshing and other reverse engineering tasks. It remains to be seen what level this will be useful.

1 Like

Ok, just looked that up on the Apple website – they have a graphic showing their lidar scan of a living room, so looks like it can do objects of larger scale as well as scanning your sneakers, mentioned in their blurb. All for a magnitude lower cost than the Leitz lidar survey equipment I looked up a while ago. This could be big. Yeah, please, McNeel go check out if the output works with Rhino.

That AR QR code was super cool :smiley:

I didn’t know anything about Sketchfab before your post. Thanks, Gustavo!

I own the same (earlier version) spokeshave. Cool you live so close to them! I made a visit to their shop about 15 years ago. Very enjoyable trip to Maine.

1 Like