I’ve just watched this video on OpenSubDiv http://www.youtube.com/watch?v=xFZazwvYc5o and I am in love.
I really hope this is on the devs wish list :eyebrowlift: (with Ptex).
This is dynamic tessellation done right, and beats the pants of the Subdiv we’ve had for all of these years.
At the 37 minute mark is what really impresses me. I remember using Maya and having to cut up Nurbs surfaces into patches manually so that deformations could be made without having to highly subdivide the whole surface. It was a pain.
Sorry if this isnt new to you guys. I just found it and was digitally aroused.
Nope. We cannot have this. At least not yet. OpenSubdiv is not ready for Blender since it doesn’t support nonmanifold meshes. Last I heard DreamWorks people were working on this, but that was quite a while ago.
And could we have two types of meshes? non-manifold and manifold meshes? or convert the standar mesh to a OpenSubd mesh… Because sometimes I will preffer have some of the good things of OpenSubd when I have modelled the object.
They “solve” this by simply not supporting nonmanifold geometry for subsurfs. Not that it’s a terribly bad solution, after all you generally don’t want nonmanifolds in your meshes. Nonetheless, Blender has always supported them since they’re sometimes useful as intermediate steps in modelling.
I don’t understand why we use the lack of support for non-manifolds as a reason not to implement OSD. You’re not supposed to model with OSD turned on anyway, as you’d constantly be rebuilding your mesh cache and as soon as you got into any kind of complex modeling you’d be waiting minutes in between each topology change. Get it working, tell people that it’s only for end of pipeline, manifold mesh display, make it clear in the wiki, and then tell anyone who complains to RTFM.
Rebuilding the FAR does take a little time, but it’s not all that much. I’m not so sure about using two different code paths for edit mode and object mode. I don’t have enough info to form an opinion on that, but it feels wrong.
But I agree, nonmanifolds are not essential. Their lack could perhaps be a little annoying sometimes, but I’d happily trade that for better subsurf performance. It’s simply the reason I’ve seen on the mailing lists and on IRC.
Why not having opensubdiv as a setting in the existing Subsurf modifier instead of replacing everything in the existing subsurf , the same as we currently have 2 settings in it ( Simple and Catmull/Clark) ?
This way if you’re doing some subsurf modelling, you’re not stuck with a non working subsurf when you have non manifold areas as can still use the catmul/clark one , then once you’re done with your subsurf modelling (and the code does not detect any non manifold area) you can switch the subsurf setting to use the pixar opensubdiv one ?
Is there a technical thing preventing opensubdiv to be implemented that way ?
That’s quite an impressive speedup, even with OpenMP and the less taxing GLSL backend.
So what would happen if you tried to change the topology with this modifier active, would the modifier auto-disable and reactivate once the operation is finished or would it disable a lot of the editmode tools? I ask because there will inevitably be a lot of people who will try to use the modifier for editing purposes (perhaps falling back to legacy and reactivating on return to object mode would work here?)
I don’t think that you can’t use opensubdiv when you modifie the mesh, you only don’t have the speed benefit that you see.
Opensubdiv build one time an array with all connected new vertice, edge, face that speed up the animation.
If you change the topologie it have to rebuild the array and that take time but not much that what we have yet.
I also want to know that… and also about the bevelled creasing we saw on opensubdiv demos that was supposed to change the modelling workflow for good (no more support loops)