Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2024 Daz Productions Inc. All Rights Reserved.
Comments
Sorry, but I disagree with that too. I believe that the Bullet sim converts everything to tris anyway. And my testing long ago showed, as I recall, it made absolutely no difference, and in fact I had a few cases where tris took a bit longer.
So, in my experience and testing, the "tris are better" and "low poly proxies are better" beliefs generally don't seem to be true. I recall even posting my results long ago, and I don't recall anyone posting results contradicting what I posted.
Again, I encourage anyone to do their own tests, and I'll be happy to change my position if the facts warrant. However, based on the personalities involved, I fully expect those beliefs to continue as folklore... :) :) :)
Can't speak to tris
- capsules are a good suggestion. I was also thinking of maybe creating spheres at the joint and attaching to the parent bone but not the child. Lots more experimentation to try.
- RE: import settings. Thanks - doh!
I did a cloth object of 13,000 quads and it took 35 seconds to run a sim for 1 second. Draped it over a sphere, small enough cloth and large enough sphere so self collisions wouldn't bog it down.
I then triangulated all of those to get 26,000 tris, and it took 35 seconds to run a sim for 1 second.
I was using the sweep hand on the Windows clock thingy, so it might be off by a second or so. But as I recalled from before, doesn't make much or any difference in my tests.
Maybe there are other variables affecting the results that say tris are faster?
Why I asked about the poly count on the figure - for the last challenge I draped a cloak over a stationary M4 .obj (unrigged) in Blender and it went very well. BTW, it draped better and faster after subdividing.
When these current testings started, I tried draping the original, undraped, cloak on the same unrigged M4 .obj in Carrara 8.1. It did the old "explode and shoot off into the yonder" that we have had since Bullett was first introduced to Carrara. I'm wondering whether there have been improvements in C8.5 that make my testing redundant?
The reason I'm doing it in Blender is because they both use Bullett and I want to test and compare - there are enough capable folk testing in Carrara:)
I did a cloth object of 13,000 quads and it took 35 seconds to run a sim for 1 second. Draped it over a sphere, small enough cloth and large enough sphere so self collisions wouldn't bog it down.
I then triangulated all of those to get 26,000 tris, and it took 35 seconds to run a sim for 1 second.
I was using the sweep hand on the Windows clock thingy, so it might be off by a second or so. But as I recalled from before, doesn't make much or any difference in my tests.
Maybe there are other variables affecting the results that say tris are faster?
Thanks for doing this but I was really talking about the stability of cloth clothing on a figure - ideally a moving figure. If you get issues with a quad mesh, and I agree that generally you will want to keep things as quads (easier to edit, cleaner mesh etc), then trying tris may help - or not - but it will be simple to try.
I remember reading Sparrowhawke's discussion when he was working on his cloth sim plugin and he was discussing things he had learned in his research. He said something to the effect that triangles required less calculation to simulate because there are only three sides instead of four. He also talked about springs and other things that allowed the flex, but at that point my eyes were glazing over and a little drool was running down my chin as my brain melted....
;-)
Anyway, it would make sense that Bullet converted to triangles for a faster sim, I mean, it was originally designed as a video game physics engine if I recall correctly.
Well, I suppose we can keep testing every different configuration we can think of to see if we find one where tris help.... :) :)
But it seems much more efficient if those who propose that tris are faster than quads can give a specific example which caused them to come to that conclusion in the first place.. :) :) :)
Do you have a specific case or example that made you come to that conclusion ? I'm a very poor mind reader... :) :)
I'm certainly not knowledgeable or an expert in cloth sims...but it could be that conversion to tris, if in fact it is done by Bullet prior to the sim, might be done to clean up some bad polygons, like non-planars. Since triangles can't be non-planar, maybe it makes the sim calculations work nicer.
Although that assumes the sim has troubles with non-planars....I'm not sure about that. But it seems clear to me that, as I've said many times before, non-uniform, nasty mesh can cause big problems in a cloth sim.
Well, I suppose we can keep testing every different configuration we can think of to see if we find one where tris help.... :) :)
But it seems much more efficient if those who propose that tris are faster than quads can give a specific example which caused them to come to that conclusion in the first place.. :) :) :)
Do you have a specific case or example that made you come to that conclusion ?
I'm confused. Didn't you, yourself, say that you thought Bullet converted to tris?
I *believe* it does, but I'm not certain. I *think* I read that somewhere long ago. And since my results using quads vs. tris have always shown no difference in the simulation times for quads vs. tris, it seems to support the idea that quads are converted to tris by Bullet anyway.
And I just posted results of a simple test that PhilW suggested that shows that there is no difference. So I'm trying to see why he believes tris are faster.
And BTW, logically, if it is believed that you can run a sim using quads, then convert those to tris and the sim will run faster, that doubles the polygon count and therefore the mesh density that the sim needs to deal with. Which is at odds with the idea that a low poly proxy object will calculate faster than a high poly one....
Well, I suppose we can keep testing every different configuration we can think of to see if we find one where tris help.... :) :)
But it seems much more efficient if those who propose that tris are faster than quads can give a specific example which caused them to come to that conclusion in the first place.. :) :) :)
Do you have a specific case or example that made you come to that conclusion ? I'm a very poor mind reader... :) :)
I was not saying that tris would run faster, I was saying that I have experience of tris being more stable than the equivalent quad mesh. In a quad mesh, each point is typically connected to four other points, and as I understand it (and again I am no expert in this), when a cloth model is run, each vertex-vertex connection is considered as a virtual spring to model the forces acting between points.
When you triangulate the model, each point is typically connected to six other points and this may be a factor in making the mesh more stable. I certainly agree that the mesh should be as regular and even as possible.
My comments on using tris came from my old experiments of trying to get a workable general cloth solution, in which at best I was only partially successful. It came when I imported clothing from Marvelous Designer and noticed that these seemed to work better than natively modelled clothes, and I was able to trace the difference to the fact that my MD imports were triangulated meshes, as that is what MD2 produces. When I tried triangulating Carrara meshes, they behaved in a more stable way, in my tests at least.
Your test with the sphere started with a solution which worked fine with quads - so there would probably be no material difference in moving to tris. All I was suggesting was, if you have issues with a standard mesh, try triangulating it and see if it makes a difference.
I wish I had the time to try some of these latest developments for myself but all of my Carrara time is given over to a commercial project that I am working on at the moment, and will be until probably April.
Just looked it up so we can know for sure. Page 32 of the Bullet Physics user manual states that soft bodies are created from "a triangle mesh". Carrara must convert the mesh to tris when the soft body modifier is added or more likely, right before the simulation runs.
Ahhh, okay...well I think your "tris might be more stable in some cases" hypothesis has developed a mythology in this forum over the years, and morphed into "tris are faster, use tris" or something like that.
Thanks for the clarification. Though I'm still not sure what the bottom line on all of this is, if there is one .... :) :)
Okay, thanks...maybe I actually read the manual on all of this? Wow, doesn't sound like me... :) :)
Diomede, got some additional questions as I'm trying to follow you step by step (please forgive the fact that I'm kind of an imbecile when it comes to modeling type stuff, as some of this may be obvious).
I've got my softbody now in place on top of my V4 for testing, did it step by step the way you mentioned, where each section was copied and pasted into the modeling room from the V4 obj that I exported and then imported into the scene, and I named each section by vertices with names that corresponded to the various body parts/bones for V4. Then sized it around V4, clicked the 'model in assembly room' and went all over the V4, using the translate function to yank out points or polys or lines or whatever, to make sure it covered V4 everywhere (although I probably didn't pull it out far enough, now that I'm thinking about it, as when V4 starts to move I know certainly muscles - thighs especially - will bulge outwards beyond their normal place In the default position).
In one of the prior instructions, you mentioned to add a general softbody modifier to the whole vertex object that's going to be the invisible underarmor, and mention using Stringtheory's settings he posted about before (with the exception that self collision is unchecked).
You then mention adding softbodies to each of the named vertices parts, but didn't give a lot of details on specific settings. Did you mean this to be the soft body attaches, not the softbody modifier itself? That's my guess, but I don't want to screw up my setup by getting this wrong as I follow in the footsteps.
Also I'm not 100% sure I'm doing the softbody attach right. I'm clicking the plus sign in modifiers to add a new soft body attach, then when that comes up, I click on the 'Edit' button and then 'Named Veritces' then select whatever specific body part of the soft underarmor I'm attaching (for example RthighS - added the S to denote it would be the softbody under armor). That lights up that section in red, showing it's selected, then I click the 'Select Tree' button and choose the bodypart bone from the V4 in the scene that corresponds (in this case, the right thigh). Is that all I need to do to attach? Or am I way off on my approach here so far. Because there are also some controls that appear in the upper left of the screen, the first one on the left being 'Validate', then 'Add' and then 'Remove' and I'm not sure what any of these do, but if I had to guess maybe I'm supposed to click the 'Validate' button to tell it that hey yeah I definitely want to attach this part of the softbody to that particular bone.
Am I doing this right, or am I way off track?
Also assuming I'm on track so far, when I add the actual cloth to the scene, how is that done? I'm guessing an ojb file of a dress from content, scaled up a little but over the V4 in the default position, then as the sim runs the dress goes back to regular scale and falls where it may and I see what happens? Just want to make sure I've got the right idea here.
All day at work I wanted to play with this more, but after work a family thing intruded and I didn't get back to my computer to mess around til late, sadly, so this might have to wait for tomorrow to actual get to the point where I'm running sims to see what happens...
Jonstark,
I have been trying to mimic what Stringtheory did with V4 here, but with other figures and other settings. (Replication is important)
http://www.daz3d.com/forums/discussion/49954/P195/#753927
That was a very important post that I had to read more than once, and I watched each video more than once.
At some point, I will try to post a true step by step tutorial (really just an example), but unfortunately I can't right now. There are at least 4 things that need to be adjusted. These include physics settings in the scene tab, the general collision setting of the base figure (V4, Aiko3, Genesis, whichever figure is being clothed), the physics modifier settings of the undersuit vertex object, and the physics modifier settings of the article of clothing to be draped.
Here are some suggestions, predominantly via Stringtheory, but everyone who has contributed to the thread as well
1) Scene tab
Under physics engine
Simulation Accuracy: 300%
Geometric Fidelity: 100%
2) Base figure (V4,...)
Select the "model" level of the figure group. Under the "effects" tab, there is a physical properties subtab. Make sure "collide with other objects" is unchecked.
3) Underarmor
a) object composition
Single object, made up of separate panels roughly corresponding to the base figure's bodyparts. It will be easier to do the softbody attach steps if each panel has had its vertices named separately. Important, for convenience of selection - should have named vertices, not just named polygons.
b) Soft Body settings (just suggestions, worth experimenting with)
Stiffness 80%
Bending 5%
Self collision: off
Margin: Checked and set to 30%
Quality: 10%
c) Soft body attaches
Separate soft body attach for each bone of the figure that you want covered in clothing (for example, maybe stomach, hip, right thigh, left thigh for a skirt for a simple figure). If the vertices of the panels have been named, then you can use the "named vertices" tab to select the panel and the "tree" tab to select the corresponding bone. Then in the upper left, click the validate button (very important). You can select the verices by using the brush and + and - tools in the upper left, but I don't recommend that. So, upper left in soft body attach room is important because you have to validate the softbody attach.
4) Article of clothing (dress, skirt, robe, or whatever)
Vertex object. Settings here will depend on what type of fabric you want (leather will not react the same way as silk, for example). So, I am less willing to put any numbers here at all.
Stiffness (vary for different fabrics)
Bending (vary for different fabrics)
Self collision: (on for something like a pleated skirt, but maybe off if shoulder or something similar could be a problem)
Margin: Checked and set to 70%
Quality: 10%
------------------------------------------------------------------------
Yes, these are the soft body attaches in 3(c) above. Yes, for each, select the vertices (easiest to use the name vertices tab, but you can use the brush and the + and - tools in the upper left). And you have to selct the bone from the tree. Then you have to validate in the upper left for each softbody attach.
Yes, the article of clothing is a new vertex object. If you already have a suitable piece of content designed for your base figure then that will be fine. Plenty of such articles are available for V4 because of Poser's dynamic cloth room has encouraged people to make dresses for V4. Marcello Teixeira has a series of free youtube tutorials if you want to make your own.
http://www.bing.com/videos/search?q=youtube+marcelo+teixeira+dress+tutorial+carrara&FORM=VIRE1#view=detail&mid=DA264CD1F2C63E8C5C72DA264CD1F2C63E8C5C72
After you click the calculate physics wand in the upper left of the assemble room, the clothing will begin draping against the softbody underarmor, which will have its panels moving along with the movement of the underlying bones of the rigged base figure.
Hope that helps.
RE: articles of clothing to be draped. You don't have to model your own if you are using V4 or other popular figures. I just went over to the renderosity free stuff area and did a search for V4 dynamic. A bunch of free meshes designed to fit V4 for use in Poser or similar programs popped up. Just note that sometimes those meshes have free floating buttons, etc. designed to be constrained in Posers cloth room, so to keep it simple, you may want to use simple articles for the moment. I suspect that Stringtheory's approach will ultimately be able to accommodate buttons, but one step at a time. I don't think I can link to renderosity. You might also find free V4 dress meshes at sharecg and similar sites.
Whoo - Hooo!!! It works, it works, IT REALLY WORKS!!! :) :)
Ok I don't think I can post a screengrab yet since I was unwise enough to do it on a naked Vicky, and even turning off the textures the nipples are pretty evident, but still...
I tried earlier in the week to make a go of this, using the method of copying all of V4's mesh and separating it out in the copy/paste method Diomede alluded to earlier. I didn't delete any of the polys though, and I found when I would run the test sims of just having V4 move (before adding any actual cloth) that for example the collar mesh does not at all correspond to the actual collarbone, and there are many parts that are like that. What I should have done is delete certain polys, stretch them/pull them into position over the actual bones, etc, but I figured I would run the sim anyway to see what happened, figuring I could always go back to the start of the sim and try again.
Even with self collision unchecked, I got weirdness, as the body parts passed through each other they would pull on each other and start to shred the 'underarmor'. I kind of expected something like this, but what surprised me is when I went back to the start of the simulation I saw some of the shredding was 'baked in' and present even at the first frame, after I cleared the simulation.
So I laid it aside for the weekend. I could see from others experiments (particularly wanted to thank Diomede who gave such clear instructions to follow) that this was working, wanted to see if I could get it working for me.
This time I just threw in one of my standard V4 characters, and inserted a vertex object, putting in a simple cylinder. I arranged it so it followed the right thigh, then pushed and pulled it to make it so it was surrounding the leg, pulling down with a bit of dynamic extrusion to make 'capules' at the top of the cylinder and rounded 'balls' at the bottom to cover the knees. I did the same for the other leg, then did 2 cylinders lower down on the two shins (now bear in mind all these cylinders, though separated from each other, were part of the same vertex object, and I named each part depending on which bone it would be riding).
That's all I did, because I expected this to fail and didn't want to spend too much time putting parts of the body all over. I put in yet another cylinder, as a separate vertex object, and sized it so that it was like a skirt, with a soft attach to the hip, the rest of it falling over the legs.
After putting soft attach for each part of the underbody, I ran the sim. Just one second, to have V4 move from her standard t pose into a running pose.
Holy crap, super fast sim! Probably less than 5 seconds. Worked perfectly, the 'skirt' turned into cloth, the under-armor softbody parts moved with their bones, no poke-through on any of the parts covered by the underbody pieces (thighs and shins) and it looked great!
There was poke through on the butt and the sides of the hip, but this was expected because I hadn't put any 'underbody' pieces there.
This really, really works, very excited! Wish I could play ;more, but have to go to a family dinner.... Hopefully later I'll be able to post screenshots that don't violate TOS in the forums, but the important thing is this works... FAST! And looks great! :)
Thank you Stringtheory for coming up with this brilliant concept! Thank you Diomede for your detailed posts and links to different parts of this discussion, made it very easy to follow along! Obviously I need to test more, but this looks like it will be a total victory in the end :) :)
Guys!! This REALLY Works! I'm so jazzed about this.
Cloth sims are fast - Really Fast! We're talking seconds, not minutes. I think it may be even faster than the dynamic hair sims I've been doing, and those are really fast too.
I am literally living the dream right now, did an animation with dynamic hair and dynamic cloth together, muhahaha. Just a proof of concept for me, nothing great to look at but the point is it really works! :) :)
Check it out as I whip V4 through a couple of random poses and see (I threw the hair on last minute as I just couldn't resist to see the 2 dynamics moving in tandem :) but that isn't at all the point)
https://www.youtube.com/watch?v=rocqEZzSgUo&feature=youtu.be
So here's my super rough setup (I really didn't expect this to work very well, or to have more problems so this is much rougher and less than precise for my component parts than I will do for a long term solution... yes my modeling skills are 'teh suck' and the mesh is pretty dang ugly. I just have 2 cyclinders for the thighs and 2 cylinders for the shins, with a little extruding and pushing and pulling to make the thighs have a rounded 'knee' section and also to push them up to cover the glutes too (I figure that thighs make glutes move more than hip does, but I could be wrongheaded about this solution, though I'll point out I get no poke through on the butt area).
Below is the Front and back views, and I put the bottoms of V4 basic wear on Vicky to preserve her modesty.
Because there is nothing else colliding against the cloth, there is a point in the animation where there is some poke through just below the navel (but above the basic wear bottoms, no nudity here, don't let it fool you).
Aside from the pokethrough below the navel, there is no pokethrough anywhere else in the animation at all. Not even at the back of the knees.
You can see the cloth itself probably could use one more level of smoothing maybe, as some of the tri's are kind of visible (or maybe that's the roughness of the soft underarmor below, not sure). Certainly now that I know this approach works I'll build a much more even and nice underarmor.
Theory: I'm thinking the thigh should include the knee area and also the glute area above. Knee could also be on the shin instead, but I'm thinking it fits more naturally with the thigh. Glutes could be attached to hip, but past experiments make me lean towards putting the glutes as part of the thigh, as the thigh seems more important to it's movement.
Also I didn't run the simulation at any kind of high accuracy, left the scene at the default 100% accuracy and both the soft underarmor and the cloth I left at 10% accuracy. Like I said, I wasn't expecting this to work well at all, it was just a rough test, I'm stunned by how well it worked. Skirt has an 80% margin, does not have self collisions checked, 35.1% stiffness, .1% bending, and 10% quality. The soft underarmor has 100% stiffness, 20% bending, 30% margin, 10% quality, and also has collisions unchecked. These aren't any kind of 'great shakes' of settings, just to let you know what I was fooling around with. Yet while there's plenty of room for improvement, you can see... it works!
Really pumped about this! Thank you StringTheory! Thanks Marcus Severus, Diomede, and everyone else who has contributed so far! Now I suspect it's just a matter of building a full softunderarmor that's not so rough and ugly, and zeroing in on the right settings!
And again, the Cloth Sims are Really Really Fast! Leaves Poser far far behind, for example. Woot! :)
PS: I have *no* clue why, but using this method the but I was encountering earlier in the thread where it wouldn't render the cloth moving seems to be gone. I know I was only like 3 people who encountered this, but this particular method also seems to overcome that bug, though I'd be curious if others who hit that bug previously will notice it's gone by following Diomede's method (which is what I basically did in this instance). PhilW, if you're out there I know you're engaged in a lengthy animation project and don't have a lot of time to play (but hey these lightning fast cloth sims might be just what the doctor ordered for an animation project, eh?) but I know you were one of the ones who hit that bug too, so thought I should mention.
Yes, I am watching this thread - can't try anything at the moment as I am rendering - animations take such a long time to render! I really need to focus on my project, as I know that if I start playing I could get side-tracked. Maybe find a little time at the weekend. Great results you are getting!
Hi Everyone,
Since before Christmas my life has been in total turmoil due to a number of things.
To make everything worse I had no telephone or Internet access for some weeks - but it hardly mattered as my computer also decided to have a major breakdown.
Anyway, now that I'm back online I've been enjoying catching up on developments here on the forum.
This thread has certainly stayed alive and some great results are coming through.
Seeing how things have progressed I took a look at one of the last attempts I made a year ago and made some slight adjustments to show what I was reduced to trying then.
This is something I showed working before (at the end of the thread last year) but didn't try to explain because it was getting altogether too weird.
But maybe the time has come to throw even something weird into the mix - who knows? It won't improve on the great strides being made by Jonstark, Diomede and others but nevertheless it may or may not help.
I'll put some images into a further post along with an explanation.
OK
first of all, if anyone has the patience for it, here is a video I posted back then:
https://www.youtube.com/embed/qLtKYZQp_Js
At the seven minute mark I showed a cube (with all its vertices selected in a soft-body-attach modifier) moving across the scene to collide with a piece of cloth which was swinging like a curtain in the wind.
At least, that's what I wanted to show. In fact the physics simulation of the cloth interfered with the path of travel of the cube. (The cube's hot-point moves towards the cloth but the cube itself 'judders' on the spot. It's like a conflict of animation methods).
Now, in the video I showed a way round this problem: I soft-body-attached the cube's vertices to a cone. When the cone moved across the screen, it dragged the cube with it - which then collided successfully with the cloth.
In the video I found a use for this technique. I was able to tuck the tail-end of some clothing into a character's sitting position.
In my experiments I tried a second use of this technique - the weird stuff.
If it's OK with everyone, I'll continue in a further post to keep my points separate from one another.
Finally, the weird stuff!
Trying to get cloth such as a dress to drape over limbs with soft body spheres/armour attached doesn't always work too well. The settings are quite difficult to get right. At its worst there will be an explosion of physics objects. At other times the limb may move as the animator intended but the sphere or capsule which is supposed to collide with the cloth becomes detached from the limb and either 'flips' outwards or hopelessly tangles with the cloth as though the cloth has too much weight for it to oppose.
The weird thing I tried to prevent this was to attach the spheres/capsules to external objects. The external objects are children of the limbs and as such move with them.
These gifs might explain things better:
The cone above the figure is a child of the figure's hip. That means it bobs along with the figure.
The cloth is soft-body-attached to that cone. The vertices around the waist were selected for that. It also keeps that part of the garment stiff.
I'm cornfused....
I thought we already determined that the idea works fine with only a few pieces of under armour. But more than that and stuff starts exploding.
I'm just trying to see where and if progress is being made. Or maybe my skull is too thick to comprehend the obvious :) :) :)
And from what I'm seeing in the demos, the collisions between cloth and under armour are fairly limited, due to the large size and stiffness of the cloth. Maybe that's why it's solving so quickly...
I dunno...I'm CORN-fused. :) :) :)
Last I played with this we were (or at least I was) looking at making under armor from the .obj mesh of the morphed character. To do that you basically export the morphed .obj, then run it thru Hex to clean up the mesh if necessary, then import the .obj back into Carrara, and it comes in with each body part a separate mesh.
Now, we also determined that if you don't need to run it thru Hex, you can just export and import, using the correct settings, and the .obj comes back in already separated by body part mesh.
Either of those methods basically gives you a duplicated and morphed (which can be very important) mesh, separated by body part.
The you just delete the parts you don't need (eyes, feet, whatever) and you have the beginnings of your under armor. See image below of the import settings.
And here is an image of the real character beside the remaining .obj body parts, ready to be made as under armor.
Again, it seems to me this method is best because it's quick and easy, AND it provides you with under armor that is already morphed to the exact size of the main character.
Now, it seems the next step is to decide if the under armor mesh is appropriate as a softbody attach mesh in a cloth solution.
Again, sorry if I'm late to the party, just trying to get the present state of affairs figured out.
Now, if you look at the hip mesh, for example, it seems that it is kinda nasty for use as a SB attach object. Maybe it's fine, but I'd want to do some testing, or clean up the mesh first, before proceeding.