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
Joe, you've done a lot of work in a very short period of time there, getting very interesting results and I wanted to say thanks for sharing your settings and method. That said, I'm using much less hardcore settings (for example in my latest sims I was playing with last night and seeing how low the settings could go before I started seeing pokethru and still getting cloth that looked and behaved like cloth) and I'm not sure they need to be set that high. I have my scene physics only at 100%, and scene collision distance all the way down to zero, and the sims are very fast but don't seem to loose any accuracy, the cloth still behaves like cloth and I'm not getting poke thru.
Roy in 8.5 when you paint the soft attach verts, they do move along with the bone you attach them too, which is good, but they are also fixed in place in relation to that bone, so for example if you wanted to soft attach some item of your clothing to your character (for example in my testing I've been using a cylinder as a skirt and soft attaching the vertexes at the very top lip of the cylinder to my character's waist) then you have to be careful where those soft attached vertexes are, because you can't just scale the cloth down and have it 'shrink' into place - the rest of the cloth will scale down but those soft attached vertexes won't move. For something like a full-body dress which relies on the fact it's falling on the shoulders to stay up and in place, then this problem won't present itself because you won't have to soft attach any portion of the cloth.
Since I was playing with a skirt, however, and wanted that top lip attached to the hip, it was tricky trying to get those vertexes exactly the distance from my V4's hip they should be before the sim starts, because the soft attach ain't gonnal let them move. If they are too far from the hip, then it looks like the skirt isn't form fitting at the hip, as there's a visible space between the cloth and the hip. If I get them too close to the soft underarmor though, it can cause intersects and tearing of the mesh of the underarmor as the sim doesn't know what to do.
Like I said, a bit tricky, but it did force me to try new scene settings, because even if it only seemed to cause little bumpiness to the 'belt' region that would not likely be noticed by anyone watching the animation, I wanted it smooth and gone and as natural as possible. So I ended up lowering the scene collision distance all the way to 0, and that helped, and also played with lowering the margin for both the soft underarmor and the cloth, and ended up with a margin of about 30% for each (I can go even lower on the cloth, down to nearly 20% and still not get pokethrough except on extreme poses).
I wanted to post an animation of my test with your low poly carrara rigged Janet character, but weirdly after the very very long simulation time it seemed to have thrown Carrara for a loop, and wouldn't let me do a render. I then thought to do screenshots at least of the various points of the animation, but when I went back to the assembly room, that instance of carrara crashed, so I got nothing to show yet, dangit. The cloth fell through the arms of the Janet figure as the sim progressed, but at the same time when I raised her knee there was no poke thru in that region, so mixed results, but the simulation for just a 2 second animation was like 6 hours or something ridiculously high, so I'm not sure what happened there but I know that's way too high of sim time to be practical, especially when I'm doing cloth sims on my softunderarmor V4 rig in just a few seconds per frame. That first test could be a one-off, I'll need to do more testing with Janet to see, hopefully a little later today even and if so I'll post results, but the long sim and the crash at the end isn't an encouraging first step, sadly :)
I'm not sure what you mean....both legs are moving, obviously, and the entire figure is in motion throughout. The under armor includes the entire body except for the arms and head, and collision with that under armor is working fine.
I think the point here is that, more than just a simple skirt, this is working with collision against an entire body, with a 7 piece under armor, and working on high density cloth that is configured to "form fit" the body, rather than just a stiff skirt colliding with 1 or 2 spheres.
In my view, this pretty much resolves the concept. I'm not sure what else needs to be proved, just refine the settings a bit. Am I missing something?
I agree, the concept definitely works, just need to dial in the settings at this point and refine to figure out what settings look like what kind of cloth, also need to find out what the best method of handling the joints will be for the softbody rig.
Btw, I think I'm changing my mind, seems to me in my testing that knees work better when connected to the shin bone rather than thigh bone, although I'm pretty certain with the right shaping of the softunderbody rig the knee section could work as part of either bone. Still clinging to the idea that the glute sections of the mesh is better off attached to the thighs than the waist. Have not done enough testing on the shoulders yet, and haven't done any testing on what will work best for the elbows, but very encouraged by the initial results that all of this is doable/solvable.
Just to say thanks for all the efforts to crack this that are being made.
Maybe later in the week I'll be freed up enough to try to build my own body-armoured figure with its own cloth.
In the meantime I played with the concept again last night with only one collision sphere attached to a cone. The cone's movement is followed by the sphere.
The cloth has a soft-body-attach region around the waist. It remains still in its place in the scene because those vertices aren't attached to any geometry. The vertices are too tightly packed, as you can see.
I played with many settings to make the cloth rubbery or flowing in different attempts. Apologies that I didn't document them. In general I reduce bending to zero and stiffness to about 28 just to begin with. I think that this one has stiffness even lower and bending about 12. But it really depends on the mesh. Self-collision of the cloth was enabled.
The simulations were very fast.
Here's a quick one I was messing around with learning how to import aniblocks and apply to V4, of course has dynamic hair too since it only took a minute to add and simulate (different sim, btw, and I had already done the cloth sim so I turned off the soft underarmor collisions with hair for the hair sim to make things go more smoothly).
Settings on the cloth are 10% Stiffness, 5% Bending (that was actually a mistake, I meant to set it lower at around 3% or below, but oh well) and 40% margin, 10% quality (and scene physics is at 0% collision distance 100% quality, and 100% geometric fidelity, additionally the invisible soft undearmor is set to 100% stiffness, 1% bending, 30% margin, and 10% quality). The cloth is simply a vertex cylinder with the top and bottom deleted, I tessellated it 1 step with midedge to center to make more squares, and then a second tessellation of vertex to center to turn it into triangles and it has a total of 2048 polygons.
The cloth sim was really fast, nearly real time actually. And render settings are low with direct lighting only to get a fast result.
The first 3 seconds she starts in a T pose, then moves to a static pose, and then moves into the aniblock, which really only starts at the 3 second mark, so it ain't the greatest animation ever, but then I'm still learning how to use these aniblock things.
Skirt is too bouncy/rubbery for my taste at these settings, but not too bad at all and it's a start at zeroing in on settings :)
Full version animation: https://www.youtube.com/watch?v=uIAp6vysYZE&feature=youtu.be
Jonstark, that is really excellent!
That, to my mind would be more than acceptable for a lot of brief shots in an animation. I like the way her hair falls over her shoulder as she looks down.
I know it looks odd and unseemly at the start, you could try flaring out your skirt/cylinder from about two thirds way down so that that lower section will form some pleats and reduce the chance of it looking rubbery. Those parts of the mesh will gather greater force as they swing down from an angle and the polygons may stretch, making the garment longer.
I generally keep cloth self-collision on. I'm not sure if that is at odds with what everyone else is doing.
second sentence edited
Thanks Marcus, I've been leaving the cloth self-collision off for a bit, since it seems to make my cloth a little less stable, but actually it's been a while since I had it checked, and now that I'm getting closer to a workable rig and some cloth settings, maybe it's time to turn self collisions back on again and see what happens...
(also yeah, I'm dynamic hair crazy at the moment, I guess I throw it into everything I do :) But it's just so damn cool, and since that hairstyle is really simple and takes less than 3 minutes to put together and only a minute or 2 to run simulations, I just can't resist adding it in)
Having stared at your work for several minutes, I have to say again - it looks great.
To think I suggested the other day that you might regret using up precious time on this! I said that after yet another long session of my own with poor results.
There is also the smoothing setting for really silky cloth. I think, but could be wrong, that it adds more vertices but not in any orderly way.
If so, I don't know what would become of a UV texture.
But that's for another day!
@joemamma
My apologies - I missed the leg bending part :)
What I'm asking is whether your settings work on a dressed model - I too get a reasonable drape using a plane, but things go pretty wonky when the model is encased in clothing.
I've been doing tests to see whether an imported .dae model works better than a native .cr2 or .duf, without the underarmour.
In the video, the first segment has v4 and Poser's Jessi, imported as .dae, with armature. In the second, it has them without armatures. All tests used the same skirt, identical settings and were simulated simultaneously. These are clearly not the software trying to make a drape, but the skirt trying to get away from it and being held back by soft body attach :)
I tried to do the same test using both .cr2's loaded from the browser, but Carrara crashed and burned!
The only real difference in the two models is that V4 is 68K and Jessi 111K polys.
This brings me to the conclusion that:-
1. Poly count does make a difference, at least in C8.1 - I might add that I also tested a lo-res James .dae in a skirt and that worked out well. To protect his reputation, he asked for it not to be shown in public!
2. The armature makes a difference - the clothing has to be "protected" from the armature.
3. Using a .dae instead of a .cr2 is far kinder to cloth sims, at least in C8.1
4. The soft body sim in C8.5 is very much improved from 8.1 and that you guys are on the right track for a workaround, at least until Daz manages to get it fully working :)
5. There really is no point in me trying to get any decent cloth sim going in C8.1
One really good discovery I've made in all of this is that users of C8.1 can get a workable Genesis2 using .dae - just have to add and skin clothing, because a clothed .dae figure brings with it an armature for each clothing item, which is not a lot of fun trying to pose :)
http://youtu.be/xHms8dfdaw4
I got a few minutes to try some stuff. I built a new underarmour based on a capsule shape with overlapping pieces for the thighs and shins, with a simple open cylinder modelled to the hip. I figured that if self-collision is turned off then that should be OK. I first tested with just that, no skirt and it seemed to work fine, following (more or less) the movements of the limbs.
Then I added a skirt, and it all exploded! Reduced the bounce and friction both to zero on the underarmour and the cloth skirt and that seemed to cure it. Playing with various settings, I was able to get a nice partial animation, with the skirt falling into pleats nicely and then following the leg up with some nice looking creases - and then it exploded. I was playing with various settings to see if I could get a complete good drape when Carrara started playing up (don't know if it was connected or not) so I saved, closed Carrara and then re-opened and loaded the scene.
But now when I try to run a simulation it just says "An error has occurred while executing" - it was working before the save, but now this. I could try deleting and re-applying all the soft body attaches, but if it needs this each time you reload, it isn't going to be a very usable simulation.
Ideally looking for answers or suggestions - but really just reporting what I found.
Wow, can't be absent from this thread for a day without missing 2 or 3 important pages.
Joe, great stuff, thanks for posting your results and settings. Very helpful. Ditto, Jonstark, Marcus, and Roy and everyone else.
Roy, the principles in C8.5 softbody don't seem to be able to translate to C8.1 . Thanks for trying it though.
PhilW - they seem to have it out or you. First the render bug, now this.;-)
RE: explosions - in addition to reducing bounce and friction settings, I have been able to address explosions by reducing the margin settings in the softbody tab of the undersuit and of the dress. Sometimes, it takes multiple reductions in margin.
No time to experiment this weekend (dang holiday-industrial-complex).
Once again I'm just popping into the forum and out again.
@Roygee, You've worked very hard, I think, to get some interesting findings for 8.1. It looks as though some things were much improved min 8.5 (eg the SB-attach vertices can attach to scene-space or to an object in the scene. when attached to an object that takes priority)
@ PhilW, perhaps if you re-open your scene you will be able to remove animation (there's a button on the SB modifier). Or maybe the error occurs to prevent opening?
@PhilW, I re-read your post and paid better attention to what you described!!
I think if you HAD removed animation before saving there might have been no problem re-loading. I can only guess that the opening error comes from the simulation-in-progress messing things up.
If you can replicate the simulation, it may be of interest to make the armour visible to see whether that is exploding. Typically, a few points of intersection with the cloth occur then it gets out of hand!
A tedious thing to try is to note that SB sliders can be moved along the time-line. It might be possible to tweak stiffness or some other setting just prior to the explosion. But, once again I come round in circles to conceding that this is very elusive and demanding.
Although some definitive basic clothing with working settings could yet be found - and Jonstark's recent example is excellent - it really may need an update from DAZ to make all the difference.
EDIT: In my last remark I was thinking how the Poser cloth room allows the user to choose what collision checks to make. You can bring vertices and faces collisions in different combinations into the simulation at a cost in speed.
Also, from observation only, when I see a Poser simulation take place frame by frame, the cloth does penetrate the body then jumps back out again before the simulation moves to the next frame. It's as though a line of code checks for that and adjusts cloth position accordingly.
I traced the issue to the soft body attach on the skirt which had mysteriously cleared all the points. I got a version "working" again, but try as I might, I always seemed to get explosions at some stage, even later stages where you would have thought that the hardest part had been resolved. I have the under-armour visible and it does seem to be that that explodes and not the actual cloth. I have spent hours doing sims today, and it still seems too unstable for reliable use. Good luck with everyone's further tests, but I can't expend any more time on it for now.
Sorry, I slept in this morning otherwise I would have been able to answer this one right away as both Diomede and I have encountered that too. Simply saving the carrara scene will not save the soft body attach setting completely correctly, and as you discovered when looking back at the soft body attaches, some of the right vertexes will be attached to the right bones, but some will just be attached to nothing or random bones. Moreover, correcting this in the assembly room is nearly impossible as it doesn't seem to want to select the right named vertexes that you named previously for each section. I've found the easiest way to correct for me was to delete all the soft attaches and then add new soft body attaches (and it would then recognized my named vertex groups, although this is only if you named them and reading back through your notes I'm seeing you went with separate capsules so may not even apply).
The way to Avoid this problem Is to not rely on the scene save at all (or maybe now that I think about it doing a different type of save, like 'save all internally' or 'save all externally' instead of the default local save might work, I'll have to test and see). But for sure a way to avoid this is to pair your character and his/her softunderarmor as a group and then drag and drop them into your objects tab for future use,, at least that's what worked for me so I can continue to re-use the softunderarmor without the tedium of group by group re-attaching the correct vertexes to the correct bones.
Philw, thanks for the tip/idea about lowering the friction and bounce on the softunderarmor and the cloth. Joe also mentioned this upthread earlier, but for some reason looking at my settings I think I must have spaced out and forgotten that the settings in the effect tab do actually matter, because I don't have them turned all the way down, and I do think that will help.
Regarding in general the 'splosions/shredding of the cloth and the soft underarmor, I believe it stems from when the vertexes of the cloth intersect with the vertexes of the underarmor. Once they hit each other, they tend to stick to each other and push/pull each other out of formation, which can cascade.
This is why it was so tricky for me working with a skirt, since I needed to soft-attach the lip of the skirt close enough to the hip that it didn't appear there was a band of air/nothing holding the skirt up. But get too close within the field of influence of the forces set up and the simulation will get confused and start intersecting vertexes (at least that's my working theory of what can cause this).
Normally the forces at play in the margin settings of both the cloth and the softarmor will work together to keep the two from getting so close as to cause problems, but scene forces and quick movements can be strong enough to still push the two together, and then once it gets 'inside the dangerzone' then the weirdness starts. I think to counter this in the scene tab we can lower the collision distance (I went all the way down to 0% and it still seems happy and much less likely to shred) which helps lower the amount of distance before the simulation trips a switch and says 'danger!' and starts getting confused about what to do. Also using as Diomede mentioned (and I believe he's absolutely right) low margin settings on both the cloth and the softarmor help lower this 'danger distance' as well, so the two items can get closer without acting up. I think it's a fine line, since if you lower the margin too much you will get some poke thru on some poses if you aren't careful (also poke through can be because the Character mesh which isn't colliding with anything will naturally deform on some movements and may in fact be moving itself outside the softarmor).
Simulations can be very fast though, I have accuracy settings of my cloth and my softarmor at only 10% and the scene tab physics I have only at 100%, which means it can be fairly quick to run a sim, see if you've got poke through, clear the animation from the cloth and softarmor, and try adjusting the margins, and running again, meaning refining the settings can be pretty quick. My theory is that once you've got the margins as low as possible without any pokethrough, you've then got it as close to the sweet spot as possible and will have little or no shredding. At that point if you want you could theoretically crank up the accuracy to run simulation (though I'm not yet really convinced how much value there is in higher values of accuracy to be honest, but I certainly would only crank it up for the 'slow final sim' before the render).
Yes this is what I'm seeing in my bullet sims in Carrara too. The character will move, and be poked through the softarmor, so quick you can barely see it, and a split instant later the softarmor will move into place on top of the mesh, in similar fashion. I imagine if we didn't have to use the underarmor we'd see the same effect with the cloth adjusting to the movements of the body, which does make sense.
Also last night before going to bed I tried checking self collisions with the exact same animation. It slowed the simulation a little, but not a huge amount, the animation really looks almost exactly the same, with a little more roughness and deformation in the self-collision version, but I'm not sure I'm seeing an improvement in folds or clothlike behavior, so I'm not sure this is a setting we'll actually need to use. I'd be all for using it anyway if I was really seeing improvements, but I actually think the animation with no self collisions looks a little better and more realistic, at least for this particular scene. As always, more testing is needed :)
Just to avoid any confusion, my underarmour was a single vertex object which consisted of five separate and overlapping pieces, hip, right and left thighs, right and left shins. I then named and softbody attached each piece to the relevant bone, with an overall soft body modifier with self collision turned off so that it ignored the overlapping - this part seemed to work just fine! It was only when I added the skirt that problems started...
Phil, your approach is very similar to my initial approach, and in practice the same as really all of the approaches, which boils down to 1 softarmor vertex object composed of many different parts, each part soft-attached to a different bone, and with self-collision disabled so the parts should theoretically pass through each other instead of distort each other as the bones move them into close proximity.
I'm finding though that to save the soft-attaches correctly, I have drop the softarmor into the objects tab, as a scene save alone will not save the soft-attaches to the correct bones.
Also just went hunting for more info on bullet. Didn't realize it was a free open source physics engine that is still being updated and refined. I'm betting the reason that Carrara 8.5 might have better and more stable results than 8.1 is it's using a more recent version of Bullet?
I also read on one of the Bullet forums that it's possible for something to be moved faster than it's sphere of influence can follow, this might go some distance to explaining some of the weird effects you can see when you are making extreme quick changes in movement. One solution might be to set higher FPS set in the animations settings maybe? (FPS = frames per second. I figure everyone in this discussion probably already knows that, but I wouldn't have, so thought I would mention).
I rather think the overall simulation accuracy (under the Physics tab) controls the size of the "steps" between calculations. 100% = one calc per frame, 200% = 2 calcs per frame, etc. So you could try increasing this if you think the speed of movement is an issue, as it will calculate smaller steps each time.
That's very interesting, and I never considered what the difference between 100% and 200% in accuracy might mean, might be very pertinent information if that's so, and it equals the number of calculations per frame.
However unless it's interjecting 'phantom' key frames in between existing keyframes, I don't know that it would be a solution to the (theoretical) problem that Bullet can have when a vertex moves too quickly in 3d space and 'outruns' it's own sphere of influence
(and bear in mind this is just theoretical that this may be what's happening, I pulled it from some discussions from the programmers over at the Bullet forums that this is a problem Bullet has currently they are trying to correct, but while it makes logical sense to my mind, I have no real idea if this might be what's contributing to the problems and freely admit I don't understand half of the technical talk they were discussing over there).
But if this is the root of the problem then I'm not sure more calculations per frame would solve anything; if the vertex is moving too far from one frame to the next and 'outruns' its own sphere of influence, then it's going to be problematic.
I'm spitballing, but seems to me if this is what's causing the issue, then if we crank up the number of frames per second on the simulation, then those in-between frames we are inserting will mean the vertex is moving less distance each time and each calculation, therefore much less likely to outrun it's own influence. It would make for a slower simulation for the same time length of the animation, but the sims are so fast that hardly matters.
And once we've got the sim done and 'baked in' to the cloth, I theorize we can just lower the frame per second rate back down to something lower so that the render can be quick too.
Just my guesstimation though, I could be way off. More testing needed (our mantra throughout this thread :) )
A quick settings note:
Was doing some simulations after turning down Bounce and Friction in the Effects tab for both the softarmor and the cloth, also had turned down the density of the cloth as well, and immediately after a few frames started getting explosions and tearing. I reset back to the defaults and isolated that I can turn the bounce down to 0% on both and things are still ok, but the friction set to 0% was the culprit for why (all other things being the same) I got nearly immediate shredding/explosions.
I don't know if the friction has to be set at it's default (85%) or not, but I can see this is an important parameter that if turned down too low will lead to explosions, wanted to throw that out there.
Another note:
I just tried the method I speculated about, which is doing a high FPS setting for the simulation and a low setting for the actual render. It works great, and I feel like an idiot for not thinking of this sooner. Since the Aniblock tutorial video suggested to use 30 FPS that's what I've been doing all along, which is higher than needed for rendering (though for cloth sims it might be advisable to crank it even higher). Especially for test animations to refine cloth settings, I don't need anything like that kind of FPS, so I just realized I can cut my render time by a third, and get faster results to refine settings. Facepalm moment for not thinking of it before. :)
Thinking along the lines of FPS and totally a shot in the dark ...years ago, when C was 32 bit and I was running it on a single-core machine, using the onboard graphics card I would often get hangups after running an animated deformer and doing a playback.
Found that by cancelling the default frame skipping, it didn't crash by letting it play back all frames. May be worth a shot to do your cloth sims with that skipping frames X disabled?
Interesting thought Roy, I'll sure give it a try. I always thought the x play had nothing to do with the sims, and was only a convenience when doing a playthrough so that the animation wouldn't take too long. I haven't had a lot of problems in my sims the way that others are reporting though, I had some explosions earlier when lowering the friction on both the softarmor and the cloth to 0%, since then I've played around and found I can lower to about 10% friction on both with no adverse effects (not sure if my mind is playing tricks, but I think it made the cloth behave more smoothly).
One other note, and this was a bit unexpected, but I thought maybe I should try adding yet another level of tessellation to take my cloth from 2000+ polys to whatever the next step would be (6000+ polys I think? I was tessellating triangles...) Anyway I thought it was worth exploring whether makng the cloth more complex would also make it behave more smoothly.
To my surprise, it was actually noticeably more rough. Still worked ok for cloth, but just looked a bit more bumpy and ridged in some places. Also the sim was a little slower, though not appreciably. I'm guessing too many moving parts is also more chances for verts to snag on each other or behave in unexpected ways. So maybe a cloth that's about 2000 polys is complex enough as it is.
Not knowing the inner workings and how various components work together, anything is worth trying :)
In the absence of any official guidelines, intensive and extensive experimentation to hit the sweet spot is the only way.
Something that may be of interest is to take a look at some of the many Blender tuts out there - although Blender has far more settings and options, they use the same Bullett engine and some of the settings are very similar - although you will have to try to understand the differences in terminology - Blender has some pretty strange terms!
Something I have noticed when doing sims in Blender is that, instead of overdoing the tessellation (subdivision), which increases poly count without smoothing, they use "subdivision surfaces", which, in Carrara terms, would be the equivalent of smoothing without converting in the VM.
Maybe try increasing that to level 2 or 3?
I believe that the following settings are irrelevant to the existing implementation of Bullet cloth:
1. Geometric Fidelity (I believe this is only used with rigid body collsions)
2. Density (as I mentioned years ago, it appears to be disconnected and moving the slider has no effect whatsoever; although I recall one beta version where it appeared to be working, long ago)
3. Bounce (I believe this too is only used with rigid body collisions)
Now I may be off on some of these, I may have missed a test along the way with one of the betas. But before you get misled into believing that a particular setting is good or bad for your results, I suggest you make sure they are even relevant for your particular simulation.
It could be a bit like saying "My car wouldn't start, until I hopped on my leg 3 times and clucked like a chicken...then it started fine" :) :) :)
Also keep in mind....
As you increase the poly count, thereby increasing the mesh density and decreasing the distance between adjacent vertices, you need to crank DOWN the margin/collision distance, or your cloth with get "rough" or even explode...
It's trying to push adjacent vertices apart until they are greater than the collision distance.
Yes, Carrara's Bullet cloth right now is too bouncy because they haven't implemented a critical feature in any cloth sim...it's called "Damping". The way cloth is modelled in sims like this is that each vertex is replaced by a weight, and each line between vertices is replaced by a spring. So your cloth becomes a big sheet of weights and springs when the cloth simulator takes over.
And as you can imagine, when you move something like that it will bounce a lot...like rubber.
Damping is inserted in there to simulate the "friction loss" that normally occurs in most fabrics as the threads of the fabric slide against each other. The result is that real cloth generally just falls and hangs without a lot of bounce. Unless of course it's heavy rubber or something.
So until DAZ connects the internal Bullet "Damping" feature, we'll have only rubber fabric :)
Now you can try the other "wind" damping (Air Drag) that simulates friction from the air as the cloth falls, but that's a bit different.
And as I mentioned, in this particular sim with SB attach, any non-zero setting of Air Drag seems to cause excessive goofiness.