At what point should I subdivide my model?
AnotherUserName
Posts: 2,726
Hi All,
So im at a point with my model where the basic shape is ready and it looks good. I need to add support mesh in several areas because I want some hard edges as opposed to the rounded surfaces I have now. What im noticing though is that as I add support mesh, im adding alot of polygons that will just be added anyways by subdivision. So, should I go ahead and add support on the basic mesh or should I just go ahead and build the support out of the subdivided mesh?
Daz 3D is part of 
Connect
DAZ Productions, Inc.
224 S 200 W, Suite #250
Salt Lake City, UT 84101
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2024 Daz Productions Inc. All Rights Reserved.
Comments
You need to have those supporting edges so subdivision won't throw it out of the shape you want. I'm not to familiar with Hex but most modeling apps work the same way. If you can I'd try toggling on and off subdivision till you get to the point that you are happy and then finalize it.
Ok, thats what ive been doing but I just wanted to be sure. Thanks Frank.
I try to save the prop before adding smoothing, then I can go back if I'm not happy with how it turns out.
What you can do (for fun purposes, the best way to learn) is to use a duplicate. Apply a level of smoothing to your object, and press CTRL+C then CTRL+D, and move the copy to one side. In the Properties tab you will see 2 objects, each with a C. In the DG click on the back arrow to remove the smoothing.
So now you have 2 objects, one smooth the other not. Whatever you do to the unsmoothed one will be mimiced on the smooth. Extrude a face, add tesselations etc and see what happens.
This all sounds very useful. Let me ask then about detailing. Is it more practical to add detailing to a model that has been subdivided or, like the support mesh, should the detailing be constructed at a base level beforehand? Lights and panelling, things like that. Also, im getting the sense that detailing is easier(?) to add as a seperate object and not as part of the main model which will eventually be grouped together with the main model.
Very hard to answer AnotherUserName, it depends what I'm making. I try do keep things as simple as possible, so I will only add smoothing when I really need it.
One of the things I tend to do is to extract the faces where I want the detail (so I am working on a copy) then hide the original. When done you just bridge things back together again. A simple example of this would be to make a cup. You would start with a circle and extrude it to make the basic shape, then copy/paste the faces where the handle would start and finish, and using those you would make the handle. From there you can bridge the two parts into one, or if you prefer just place it into position. The latter is sometimes a better route because you may have added more tesselations which would spoil original.
Also, bear in mind the materials, with the cup example - the cup would be glossy while the coffee would be a liquid. I realise it can be done with seperate maps but why make life hard for yourself.
That makes sense John. I think I have a better idea of how to proceed. Putting a model together is quite a puzzle. Its much more challenging than I thought it would be, in a good way though.
First, let's talk a bit about subdivision. Most people look at the various sub-d levels present on the right side of Hexagon's UI and think "that's all there is." Well.... You can actually go into the Surface Modeling tab, pick the "Smooth" (Which is really Subdivision and I don't know why they choose to call it "Smooth(ing") and choose the type of sub-d algorithm you want to use. ALSO, and this is important for your application, you can choose "break points" using this tool, which allows you to preserve edges. You may want to check those tools, regarding your post above. (You'd likely use Catmull-Clarke subdivsion, since it produces good, clean, quads and you can set break points when using it. It's the same sub-d algorithm that Hexagon uses for its "smoothing levels" tool in the main UI.)
Only choose to Sub-D those parts of your model that actually benefit from having a higher mesh density or benefit from the addition of extra geometry and the "smoothing" function that the chosen algorithm produces. Don't Sub-D willy-nilly or sub-d your entire model if it's not necessary. In general, most forms of Sub-D are not desirable for inorganic, hard-surface, models with lots of clean, crisp, edges. While Hexagon does allow the use of Break points in various sub-d algorithms, the very nature of those algorithm will tend to produce soft-edged surfaces and, with Break points, sometimes unwanted "puckers" can develop as the algorithm tries to only minimally interact with the vertices in those areas. When in doubt, test extensively and don't create extra geometry you don't need.
And, it's all in the details... Don't try to create one contiguous mesh if you absolutely do not need it. For one, it's extremely difficult and messy to do with a model that has large surface areas and bits and pieces of detail scattered about. For another, some of the topology gimcrackery you'd be forced to use in order to create certain sorts of details may react strangely with your chosen renderer. For 3D still rendering, most renderers work well with all quads. Game engines use tris, because it's fewer verts per face to run calcutions for, not because the model size is necessarily any smaller. (Though, there are some advantages there, depending on the type of model.)
Look at your model. What do you absolutely NEED to have built into one, contiguous, surface? It's not likely to be much. OK, so then you think about what topological features make the most sense to add to the main body of the model and which ones make the most sense, from the perspective of mesh weight, actualy usage of the model, the renderer, the all important "editing" consideration and texturing demands, to be modeled as a separate mesh/group.
Let's look at a model..
http://www.matus1976.com/3d/sd/images/star_destroyer_5.jpg
So, when planning out that Star Destroyer, what would your separate groups look like? Well, you'd likely have the main, triangular, shape as one group. Perhaps, it'd be best to have in a clamshell arrangement. There'd be the main body top, a separate group for the sides, the bottom and then a separate group for the back. All of these groups would have minimum details, all of them being gross structural details with no greebles/finicky bits. See the "raised plate" features on the main body? Not the superstructure, but those big armor-looking plates? Those would be separated from the main body. Why? Because - Adding those as part of the main structure would add more verts than needed, especially if you're going for all quads. Also, you'd have to put in extra edge loops for renderers like DS and Poser, since they don't do hard edges will without them, anyway. AND, if you're planning on soft-shading in your renderer, you'll be adding even more, since the renderer will likely make a hash of your hard edges without them if you did one contiguous group for just that main body, alone.
Look at the superstructure. How many seperate object meshes would you use for the basic, blocky, shapes? I count five basic boxlike shapes. Can any be combined? Maybe two. Do I "absolutely need" to combine them? No. And, if I did, it'd just make for a heavier mesh that served no useful purpose for its extra weight. Since I can have objects intersect in both DS and Poser, if I were modeling it for those apps, they'd all be seperate box-like structures that intersected eachother, with the first level instersecting the body.
What about the conning tower and bridge area? Two separate meshes, IMO. The engines/cowling? One model each of both engine types, then duped and moved as needed. I'd also make a seperate stern cap.
Then, there's the details. First, we'd do a couple of different laser canon types and dupe/place those as needed. Then, the bridge details and shield generator. I'd also add some primitive hanger bays set up to receive a matte texture so they look like they had activity/ships/personnel in them or something.
All in all, there'd likely be over twenty different "groups" of non-contiguous geometry and that's without the greebles and without counting duplicated geometry. (Greebles - Little fiddly bits of geometry that I can't reproduce with normal/bump/displacement maps that add nice surface details and keep the eye and imagination "active" as they roam over the surface of the model.)
And that also means that UV'ing this model would be much, much, much easier. Some groups could be mapped solely using a primitive projection, without any touch-up. Some would probably need to be "unfolded" with some seams manually added to give the best UV map, though.
A general way to go about considering how to judge how you should plan groups of geometry, off the top of my head:
1) Will geometry be duped/cut/pasted? (Like with the two engine types and laser cannons)
2) What parts of the model can be modeled in a symmetrical state, using mirroring to reduce workload and improve quality?
3) What options present me with the cleanest possible topology?
4) What options present me with the best ways to UV map the model? (Something you need to plan for in Hexagon, specifically, 'cause it can be finicky at times.)
5) What "shapes" do I see? How many of them are "primitive-like" shapes and how many of them require asymmetry or details that are outside of the main "form" of the model. (ie: The main genre of "form" for the model would be either "Inorganic" or "Organic." Inorganic shapes, like hard edges, boxes, geometrical shapes vs Organic shapes, like a human elbow, face or an octopus a cat a cow a slime-mold, etc...)
6) Which of the combinations of shapes are easily combined, like related geometric shapes with similar, close-by, geometric shapes, without much fuss? Which shapes can't easily be combined with others? (This is mostly an inorganic model. But, it has widely varying bits of details and differing types of generally primitive geometry. However, if I wanted to stick a squid's head on it, coming out of the conning tower, then I'd be combining not only general geometric families, but I'd be forced to create a lot of mesh density in very simple areas of the model, density that is absolutely not necessary if I only modeled the conning tower and squid head separetely, as individual meshes/groups.)
7) Details - Which details require more density of geometry for their area on the model than the surrounding area of the detail requires? (ie: A cylinder shape that sits on top of a box. The cylinder requires lots of geomtry for a nice curved surface for the tube. The box only requires six faces or five, if one face can be removed and then that box intersected with another surface. Making those two, separate, meshes makes sense. Making that combined shape one contiguous surface doesn't make much sense from this particular perspective.)
See the theme, there? Plan in order to make your mesh the simplest and less cluttered mesh possible for the model that you're building. Take certain considerations into account, like the demands of the application the model is being used in and what sort of texturing/mapping you need to do. Take into account presentation and versatility, as well. For intance, I planned for boxes to be modeled for the visible hangers and mapped so that a background "matte" could be used to simulate activity going on in the hanger, for low-resolution/long-distance renders. That's a few extra faces, but it's not very many and it increases the versatility/presentation of the overall model a great deal.
PS - A special consideration for Hexagon: UV-Mapping
Hexagon UV Maps best when it is dealing with symmetrical geometry. That goes for both organic and inorganic shapes. With a symmetrical object, Hexagon does a good job of giving a clean UV Map, no matter if you're using projections or unwrapping it. When you're planning a model, always place a prime importance on symmetrical attributes when you're using Hexagon to generate your UV Maps. When in doubt, choose to use symmetry function as much as possible. Check your symmetry, from time to time, by cutting/copy and pasting half the model, using the "Mirror" function and then rewelding it all back together along the axis of symmetry. Seriously - Do this. :) You'll be thankful that you paid attention to those bits of your model that can be symmetrically modeled and UV Mapped, seperately, away from the counfounding geometry of asymmetrical bits.
This is great information. Thanks Morkonan. I think that using break points is going to save me from alot of grief. Might take a little practice...
Im anxious to post some pics of the mesh so that I can get some critiques. I was hoping to do that relatively soon but im going to go back and do some revision now that I have more info.
Cool! It'd be great to see your progression. Just a cautionary note: Always use multiple revision level save files. ie: Object1_Stage1.hxn, Object1_Stage2.hxn, etc.. That way, you can always go back to an earlier version, no matter if you bork up your base mesh in the current save or not. Also, always save an un-subdivided version of your model in your working file. (Hide it when necessary) Always. Always.. I still remember a huge model I did and then forgot to save it in its unsubdivided state. That was scores of hours of work, at least, completely borked and wasted, since I ended up having to go back and revise the model... in its subdivided state. I was stubborn - I could have saved time just reproducing a whole new low-density version from scratch and going on from there. :) But, if I had a nice clean backup of it in its raw form, I could have fixed the issue in minutes and been on my way.
On a tech note - Disable autosave in Hex. It's unreliable, just in case you're wondering, and can cause crashes. Far better to just practice good revision/save discipline. Your mileage may vary, though. It's just a passing thought.
Good luck!