🧵 Shrinkwrap preserve volume
Anonymous at Mon, 4 Dec 2023 07:32:32 UTC No. 966092
I've stumbled upon an interesting mechanic involving the Shrikwrap modifier. Maybe it's obvious to some but I've never seen it mentioned before, so I'll show you the thing first and I'll attempt to explain what's going on later.
Anonymous at Mon, 4 Dec 2023 07:44:12 UTC No. 966093
>>966092
>blender 2.49
forced soul
Anonymous at Mon, 4 Dec 2023 07:57:42 UTC No. 966094
If you compare the quality of the same thing made with the Softbody simulator, it's a lot better. Mostly because the simulator is able to pull on distant vertices and bring them closer to the bend.
Anonymous at Mon, 4 Dec 2023 08:08:55 UTC No. 966095
>>966093
I've made a patch, it's here:
https://sourceforge.net/projects/sn
Apply it to:
https://download.blender.org/source
Make sure Python 2 is installed and recompile. It's not bad. Back in the days we used to complain a lot about the UI but we had not idea it was about to get a lot worse.
Anonymous at Mon, 4 Dec 2023 09:25:05 UTC No. 966099
If I was a real clay sculptor (which I'm not but I've watched a youtube tutorial once, so I'm basically an expert) I would know that in order to model anything I would need a setup similar to pic related.
One reason is that clay shrinks a little when it dries, so a solid ball of clay would crack. In fact I've seen artists put all sorts of junk inside clay sculptures. Junk such as light bulbs, plastic toys and who knows what else.
The other reason is volume preservation. As I imagine my final shape, I can add balls of newspaper to block off volumes which I'm sure my final surface will not intersect.
🗑️ Anonymous at Mon, 4 Dec 2023 11:19:32 UTC No. 966105
>>966099
You do that because newspaper and wire and little bits of wood are cheaper than clay and a lot tighter too. Also if you have stuff sticking out like arms clay would just collapse so you need an armature.
You can make a solid clay model, if you dry it slowly, then the cracks will be smaller and you can fill them in with clay.
Clay for a classical sculptor is usually used to create a sketch that you then transfer into stone or wood with a proportional caliper.
Most clay you get is meant to make pots, sculptors clay should have a dark color, you can get a dark grey or a dark green clay without any chamotte in it.
It fucking sucks to work with the shitty red clay meant for pots and bricks, it's horrible to see shapes in it compared to a dark grey/green one.
Anonymous at Mon, 4 Dec 2023 12:04:33 UTC No. 966107
>>966099
No, the art of sculpting is detailed and carefully constructed model. If your model is deformed then that’s a problem with you, stop adding extra clay. Nothing on YouTube can teach you how sculpting works, people have done it without putting anything inside the model. People made sculptures because they wanted to and not because they followed some kind of popular people.
Anonymous at Mon, 4 Dec 2023 14:02:50 UTC No. 966115
>>966107
You wanted to sound really smart but you ended up sounding retarded
Anonymous at Mon, 4 Dec 2023 15:25:34 UTC No. 966123
Bending stuff using the Armature system causes models to lose volume. If we add some inner object centered at the joint but below the surface of the main model, it'll be invisible but as the bend progresses, it'll show up more and more.
The Shrinkwrap modifier notices that some vertices from the main model are inside the inner object, so it re-projects them on the surface of the inner object and places them there.
I've been doing something similar using the Softbody simulator for a while now and I'm getting really good results (some of which I've already shown months ago) but people kept complaining to me because simulations are bad because they have numbers and stuff, so here's a cheaper (somewhat lower quality) alternative based on the Shrinkwrap modifier.
Anonymous at Mon, 4 Dec 2023 15:32:39 UTC No. 966124
>>966092
https://www.youtube.com/watch?v=1Wm
Anonymous at Mon, 4 Dec 2023 15:43:57 UTC No. 966126
>>966123
What codec settings did you use to make that webm?
Curious because recently more and more webms show up that won't play in my browser while most other webms play just fine.
Anonymous at Mon, 4 Dec 2023 15:47:23 UTC No. 966127
Besides bending, as I've mentioned here >>966099 I would like to know if something like this could help me with sculpting.
From the few attempts I've made to sculpt in Blender I've noticed that the Smooth tool suffers from the same volume loss issues that the Armature system has. It tries to linearly interpolate over a curved surface and I don't like the result. Also, without some reference volume like the balls of newspaper would give me in reality I tend to get lost.
But this is just as idea for now and I'll explore it a little bit a the time.
Anonymous at Mon, 4 Dec 2023 15:48:49 UTC No. 966128
>>966126
I typed this command line:
ffmpeg -i /tmp/0001-0100.avi 0001-0100.webm
and posted it.
Anonymous at Mon, 4 Dec 2023 16:06:56 UTC No. 966129
>>966092
>>966123
What is the benefit here over just doing corrective shapes or using corrective joints? Especially the elbow/knee is probably one of the easiest things to fix wit those considering the limited rotation they need to do. I don't know about blender, but at least in Maya, while shrinkwrap is nice in some situation, it kills performance, at least on anything with a slightly higher poly count. Basically, for the cost of a single shrinkwrap, the scene could instead be running 3 active character rigs at the same time for less of a performance impact. It has its uses of course, but I'd certainly not build a rig that heavily relies on it.
Anonymous at Mon, 4 Dec 2023 16:32:39 UTC No. 966130
>>966129
You can use both the Softbody simulator and the Shrinkwrap modifier to create Shape Keys if you want to, for example, to export outside of Blender.
I make short rendered animations, so Shape Keys are not a concern to me because simulations look so much better in this context.
Anonymous at Mon, 4 Dec 2023 18:43:52 UTC No. 966141
>>966094
>>966123
We've had this discussion over and over months ago, just stop.
Your method produces extremely obvious lowpoly geometry on the outside of the bend.
Anonymous at Mon, 4 Dec 2023 20:42:43 UTC No. 966148
>>966141
Look at the top of the screen where it says:
Ve:2634 Fa:2842
It's a low poly not subdivided shape with a 130 degrees angle bend. It's good in both cases. Also since last year I've presented four completely different methods and I have even another one which I'll present shortly.
Anonymous at Mon, 4 Dec 2023 21:39:46 UTC No. 966151
I was looking at AccuRIG because I wanted to know if they had any gimmick to improve the bends. Mixamo doesn't have any.
They call it "Partitioned Skin Weight for Natural Joint Bending". If somebody here owns a Windows PC and would like to investigate I would be curious to know what that is.
https://www.youtube.com/watch?v=JK3
I suspect they mark some areas of the model in such a way that the number of influences is limited to one or two, but I don't know for sure.
Anonymous at Tue, 5 Dec 2023 10:08:05 UTC No. 966190
>>966151
>If somebody here owns a Windows PC and would like to investigate I would be curious to know what that is.
I doubt anyone will be able to deduce their secret sauce from using the software.
I'll try it out, but aren't you the anon that posted about how you've solved that exact thing already in your blender fork?
Anonymous at Tue, 5 Dec 2023 11:04:39 UTC No. 966192
>>966190
Yes. I proposed several solutions of my own some involving patches to Bender, some not, but it's a subject 'm interested in a general sense.
For example I was looking at this the other day:
https://smpl.is.tue.mpg.de/index.ht
Basically they derived about 200 general purpose Shape Keys using machine learning and a Python script to apply them from a large set of 3D scans. I would never use something like that, but I research and watch it.
🗑️ Anonymous at Tue, 5 Dec 2023 11:06:17 UTC No. 966193
>>966151
https://hal.science/hal-00407571/do
Anonymous at Tue, 5 Dec 2023 11:10:19 UTC No. 966194
https://www.youtube.com/watch?v=2s3
Anonymous at Tue, 5 Dec 2023 13:10:28 UTC No. 966199
Hey, you can check this too if dual quaternion isn't good enough.
https://apps.autodesk.com/MAYA/en/D
Anonymous at Tue, 5 Dec 2023 14:57:48 UTC No. 966203
>>966199
The issue with Dual Quaternion or Preserve Volume as it's called in recent Blender is that while the "normal" way to interpolate is guaranteed to lose volume, the Dual Quaternion method is guaranteed to gain volume.
In other words, it has the same problem except in the opposite direction. For a shape to gain volume is even more un-natural than for a shape to lose some volume. Dual Quaternion is also a lot more expensive to compute, which is why game engines don't implement it at all.
Anonymous at Tue, 5 Dec 2023 14:59:08 UTC No. 966204
>>966192
>https://smpl.is.tue.mpg.de/index.h
Incredible research.
From what I gather you can only use their pretrained model with their basemeshes.
They also seem to have some restrictions on who they give their full version to.
I wonder if there's enough resources to train your own version for your mesh. They say in the video they've used public datasets, and they have a github.
Need to investigate that...
Anonymous at Tue, 5 Dec 2023 15:11:34 UTC No. 966205
>>966199
So, you may say: "but wait a minute... if normal interpolation gives me an underestimate of the volume and Dual Quaternion gives me an overestimate, can't I just average between the two?"
If look at my modifier stack you'll see that it's possible to apply the Armature modifier twice. In addition you can control how much of one or the other via a vertex group.
Anonymous at Tue, 5 Dec 2023 15:21:28 UTC No. 966206
Are you that schizo shoulder volume preserve guy? You're not going to find a "hack" that the rest of the cg community (blender, auto, side) hasn't come up with yet. Just give up already. This is heartbreaking.
Anonymous at Tue, 5 Dec 2023 15:22:13 UTC No. 966207
>>966204
The first half of this presentation by them:
https://www.youtube.com/watch?v=rzp
is a very interesting historical overview. The second half is more specific and they talk about the datasets.
And at the very end they mention Makehuman and Microsoft Rocketbox >>966174 That's where I've found out about it.
Anonymous at Tue, 5 Dec 2023 15:28:17 UTC No. 966208
>>966203
You have a couple of papers about enhanced Dual Quaternion in Scholar, I think the solution is that, also
> game engines don't implement it at all.
https://godotengine.org/asset-libra
Anonymous at Tue, 5 Dec 2023 15:31:40 UTC No. 966210
https://gist.github.com/maxattack/6
https://www.daz3d.com/forums/discus
Anonymous at Tue, 5 Dec 2023 15:31:53 UTC No. 966211
>>966206
I'm not a the shoulders guy and 3D model bending is an active area of research. Go back to model some anime figurine and don't worry about me.
Anonymous at Tue, 5 Dec 2023 15:34:28 UTC No. 966212
>>966208
If he was able to implement Dual Quaternion in GLSL, that's kind of interesting. I didn't know it was possible but still, you have to admit it's a rarity.
Anonymous at Tue, 5 Dec 2023 15:38:05 UTC No. 966213
>>966211
If you want to preserve volume, you use Machine Learning to override traditional skinning, like SideFX introduced in Houdini 20. Make a few ground truth poses and then train them with pytorch. Don't be a complete idiot.
Anonymous at Wed, 6 Dec 2023 14:16:53 UTC No. 966326
take a look at this.
https://rodolphe-vaillant.fr/entry/
https://www.youtube.com/watch?v=GyO
Anonymous at Wed, 6 Dec 2023 15:36:59 UTC No. 966335
>>966326
I saw that a while ago. You can get very close to those results using the Cloth simulator with self-intersection enabled and a lot of patience to find the right parameters. I don't a have a demo off hand to show you but I did post some.
I haven't investigated it much because I chose to focus on simulations because I wanted support for dynamic effects and collisions with foreign objects (yes, for the reason you can imagine).
If they're able to that in real time without a simulation it's really impressive. I'm curious to know how complex is the code behind it is.
Anonymous at Wed, 6 Dec 2023 15:49:40 UTC No. 966338
>>966335
idk, I've found it on projects.blender with the Anti Bulging ... patch?, ofc nobody is working on that.
https://projects.blender.org/blende
🗑️ Anonymous at Wed, 6 Dec 2023 18:48:49 UTC No. 966348
>>966335
more about the topic
https://github.com/pmbittner/Optimi
https://github.com/pmbittner/Optimi
🗑️ Anonymous at Wed, 6 Dec 2023 18:50:13 UTC No. 966349
>>966335
more about the topic
https://github.com/pmbittner/Optimi
https://www.youtube.com/watch?v=DfI
Anonymous at Wed, 6 Dec 2023 18:56:54 UTC No. 966350
>>966335
more about the topic:
https://rodolphe-vaillant.fr/entry/
and a different aproach
https://github.com/pmbittner/Optimi
https://www.youtube.com/watch?v=DfI
Anonymous at Wed, 6 Dec 2023 23:07:52 UTC No. 966369
>>966350
"Real-time Skeletal Skinning with Optimized Centers of Rotation" is an interesting concept but doesn't look like it's free to use or implement.
Let me show where I'm at on my own:
https://www.youtube.com/watch?v=n1p
The second half of the video is in wireframe, so you can see how my rig works. I can't show you the foreign objects collision functionality, but it's there.
I don't think the concept is radically different from the "Elastic Implicit Skinning" thing, except I'm using an explicit solver. My simulations run at 3 to 6 fps on a good day on a fullbody model without self collision, they claim 96 fps and even GPU implementation.
The other methods (the ones that don't involve some sort of simulation) to me don't look radically different than this >>966205 which I've posted because I wanted to show that it was available at least since Blender 2.49
Anonymous at Fri, 8 Dec 2023 00:42:31 UTC No. 966462
Before I forget, for the two people in the word who may be interested in programming and simulations, a couple things I've stumbled upon.
https://github.com/mmmovania/opencl
These are reference implementations of a generic piece of cloth with several different solvers. I was able to port the Verlet integrator to Blender's Softbody simulator, which is what I've used to make >>966350
My goal would be to eventually port over the implicit integrator but right now I'm having an hard time doing it for a very specific reason. If anybody here is really good at math I can try to explain it.
https://github.com/BrunoLevy/geogra
Up to Blender 2.79 the bone heat algorithm depended on OpenNL, since then it's not n blender anymore but it evolved into this. There's really advanced stuff in it such as surface reconstruction from point clouds. There's also a method to break down volumes into tetrahedrons, which I think could be useful all sorts of quality volumetric effects.
Anonymous at Fri, 8 Dec 2023 00:48:07 UTC No. 966463
In >>966462 I meant to reference >>966369 not >>966350 Sorry about that.
Anonymous at Sat, 30 Dec 2023 00:24:17 UTC No. 968797
>>966092
You are going to enjoy this.
https://www.youtube.com/watch?v=_kj
Autist creates a entirely procedural basemesh with geonodes then uses drivers from armature posebones to fix skin self-intersection extreme joint bends.
It's not entirely clear what he does and he rambles on quite a bit but ngl it looks like the most viable approach to the problem you can do without modfiying the blender source code.
Anonymous at Sat, 30 Dec 2023 01:58:24 UTC No. 968800
>>968797
you lose mass with no bulging on intersection. This is horrible
Anonymous at Sat, 30 Dec 2023 03:21:28 UTC No. 968804
>>968797
Thank you. I saw that on Blenderartists some time ago. I thin he posted in the "shoulder guy"'s thread over there even.
The problem is that some people treat Geometry Nodes like it was a game of slot machines. They put together some spaghetti monster, they pull the handle and they conclude that they're "really close".
Anonymous at Sat, 30 Dec 2023 04:59:16 UTC No. 968807
If you share my interest in physics simulations I've found a really nice video today:
https://www.youtube.com/watch?v=-GW
It's nothing special but it explains in detail and with simple code how to do nonsense like pic related.
Just in case you're tired of making spaghetti monsters in Geometry Nodes and want try something else. Just saying.
Anonymous at Sat, 30 Dec 2023 13:32:07 UTC No. 968831
>>968804
>The problem is that some people treat Geometry Nodes like it was a game of slot machines. They put together some spaghetti monster, they pull the handle and they conclude that they're "really close".
I agree.
But that has more to do with people going into it with insufficient knowledge of what each node can do partly due to poor documentation partly because laziness.
You can have this with written code too and as a counter example: Circuit diagrams are done using "nodes" and there the attitude is quite rigorous, better than say writing python script or game engine source code.
>>968800
Preserving mass is generally something that's difficult to do on a mesh working only with surface data. Brute force approaches like voxelisation are possible but too slow to work with.
But perhaps there's a way by normalizing the offset amount locally.
Volume preservation will generally be a separate process when working with surface meshes as opposed to a representation that's already volumetric.
Anonymous at Sat, 30 Dec 2023 14:48:34 UTC No. 968838
>>968797
>without modfiying the blender source code
This >>966791 is a combination of >>966205
and the Cloth simulator. Unfortunately nobody has worked on it since the 2.9x series and it's slow and it has problems.
>>968831
I've talked about realtime softbody physics here >>966268 It's starting to become available almost anywhere in game engines. Maybe it'll become super popular this year after they release the game.
>Circuit diagrams are done using "nodes"
I've done that a lot with my soldering iron. That's where I've learned that either you know the basic circuits, or if you try to go by feeling you're going to get nowhere.