Every good Catmull Clark Subdivision Surface implementation with Edge Weighting gives you perfect visual continuity in reflections.
The mushyness is countered by the edge-weighting.
You have other problems with SDS that are easier to solve with NURBS, but for relatively organic forms it's perfect.

For example:
http://www.screendream.de/Site/3d-design/sculpture.html#/3

That is from Softimage XSI, which has one of the best implementations I've seen. After Autodesk bought them, I ended my subscription when it became clear that it is no longer going anywhere, but I still have a license. Now I work more in Lightwave 3D again, where the SDS and especially the edge-handling aren't as good, but still you can get similar results, although with a bit more fiddling.

Since I come from the polygon-world originally, I'm always amazed how much effort and money are needed to get something similar in NURBS. ;-)

Cheers,

Tom

EDIT: Just to be clear: what I am looking for is visual continuity in reflections. The discussions aren't always clear if G2 or G3 is needed for that.
Sorry for any confusion that may have caused.

Hi Tom, you only need G2 for there to be visual continuity with reflections, G3 is more for bragging rights than anything you can actually see on the geometry.

- Michael

hehehe - okay, I take your word for it.

I usually use G3 in MoI when it works since I like the created profile - looks more natural or organic to me, similar to what SDS create, a certain kind of flow. Most of the time NURBS look too dead to me.

Cheers,

Tom

Hi Tom, yeah if you're trying to make organic looking forms, usually sub-d modeling is a better fit for that style of stuff than NURBS modeling.

NURBS modeling is good for shapes that are well defined by 2D profile curves.

- Michael

""""""""""Every good Catmull Clark Subdivision Surface implementation with Edge Weighting gives you perfect visual continuity in reflections.
The mushyness is countered by the edge-weighting. """"""""""

Catmull clark is "trying" to give you C2, though it can still result in C1 at extraordinary vertices. It does this by melting everything down to try that average. Repeated subs can get close. The "perfect VISUAL curvature" result just means you "cant see the difference".

I was thrown by the request being "G3", or even "G2". There's is a difference.

"""""""""Since I come from the polygon-world originally, I'm always amazed how much effort and money are needed to get something similar in NURBS"""""""""

That's because the NURBS are actually perfect, and "DO" represent that. It's much easier to kindof "average everything down" to be somewhat close........

But like Michael points out, YOU cant really see the difference.

But there is a need in very specialized situations, where those values cant just be "moved around"........

Well, since it seems that there's no G2 Fillets in any affordable software either, it's moot anyway. ;-)

And I have little influence on what my customers choose to do their work in. ;-)

Let's leave it at that.

Cheers,

Tom

"""""""""Well, since it seems that there's no G2 Fillets in any affordable software either, it's moot anyway. """"""""""""""

Well, MoI does G2 fillets. Also G3 fillets as well.

""""""""""It's really too bad that on these models, imported from Catia as Step, I can't even fillet a single trivial edge, no matter what mode in MoI."""""""

But why?

Can you produce "ANY" example that doesn't work, that is not under NDA? I mean, a "single trivial edge" is kindof a broad, general statement.

Burr, let it rest, it's okay.

Wow. Ok Tom.

Darn! I wanted to see these 'unfilletable' edges too, oh well I guess we'll just stick with edges that do fillet.

-

Just to add some follow up on this:
Catia is unable to fillet those models as well, neither me nor my customer really understand why, but before we dive into endless debugging sessions, we decided that I'll rebuild the models for now as SDS objects - and he will look into MoI/Rhino for future construction. :-)

I only need the fillets for visualisation and rendering, so the results from SDS are just fine, therefore I didn't want to go into a loop discussion ;-)
Sorry if I came over defensive.

Thank you all very much for the input!

Cheers,

Tom