Unreal Engine how can we achieve soft Skin Physics in Unreal Engine

[mikeblack]

It ain't that hard to get accurate location, i've done it before here, it's not exactly an elegant solution though but I imagine yuo must be doing something relatively similar. It's not without it's problems though, i think if I revisited this I would try to move the semen projectile from the niagara emitter to being just a normal physical mesh because the niagara collisions tend not to be very reliable, especially for gpu particles., i think for the regular collision approach on gpu particles thhey only collide if they are rendered, and they only render if they are not obscured by anything (such as the mesh or even another particle)

It is a problem I could solve with scratch, but I think just moving the projectile itself to a blueprint actor and forwarding it's collision & position data to the niagara emitter would be simpler at the end of the day.

It's also possible to use CPU particles for the projectile, but to forward data from CPU particles to GPU particles the only way i know is to send the data from the cpu particle to a blueprint actor and then send it back from the blueprint actor to the GPU particle emitter which is a mite convoluted.

Interesting so you spawn particles for all triangles and check how close they are to get the one with the correct triangle id. I'm trying to do the same just without the spawn a particle for every tirangle part. With a roundtrip to blueprint to figure out the triangle id of the hit there and pass it back to niagara.

Looks like it has potential. I use

You must be registered to see the links

plugin to handle any kind of data retrieval when tracing against a skeletal mesh. Can get the hit UVs, triangle, vertex position etc. It's super handy.

Looks useful. Will try to get the triangle id of the hit with it.

 

[Velomous]

Interesting so you spawn particles for all triangles and check how close they are to get the one with the correct triangle id. I'm trying to do the same just without the spawn a particle for every tirangle part. With a roundtrip to blueprint to figure out the triangle id of the hit there and pass it back to niagara.

Yeah, like I said, not elegant. Do let me know if you figure it out!

 

[mikeblack]

Do let me know if you figure it out!

It works with the vertex plugin but has it's own downsides with my current implementation.
Preparation - create mapping from vertex index to triangle index:
1. In Niagara spawn a particle for each triangle location of the skeletal mesh (like in your setup).
2. Send all triangle positions + indexes to blueprint (had to limit here how many it sends at one time due to engine restrictions of around 32k active events).
3. Use vertex paint plugin "Get Closest Vertex Data on Mesh" function. Either can have it find the closest vertex for each triangle position one at a time or use it to get all vertex positions directly and do a manual closest distance matching. For each vertex index save the index of the closest triangle in a array.
4. Save the array to file.

Actual use:
1. Load the file to array being a mapping of vertex index to index of closest triangle.
2. Spawn collider.
3. On collision event run "Get Closest Vertex Data on Mesh" to receive closest vertex index.
4. With the array from step 1 map it to it's matching triangle index.
5. Send triangle index to Niagara for effect.

Downsides:
1. Looses some precision. The vertex count is around half of the triangles so mapping from vertex won't give the exact triangle at hit location.

Vertex res:

Triangle res:

2. The roundtrip takes quite a bit time.
Niagara CPU hit event to blueprint->blueprint closest vertex check until event->send back to Niagara GPU
has a visible delay. Not so much regarding performance as it runs via events but each step does add a bit delay until events are received. The closest vertex check also says it takes around 25ms. It all adds up creating a visible delay.
Didn't try much optimization yet. Like the closest vertex check would allow to pass the bone of the hit to improve performance there which I didn't use.

 

[razfaz]

lol Poor Manequin. He/She has to endure a lot.

PS: Niagara is fun to play with.

 

[Velomous]

Well guys, looks like the holy grail might have been found for us.


Of course, like always, the video says virtually nothing about how it's done. Frustrating shit.

This is what it says:

This is my first quick animation done using new skeletal mesh deformers in Unreal Engine 5.5! I tested every one of three available deformers on a MetaHuman face, added face animation recorded with MetaHuman Animator and animated the body and Mannequin hands by hand.

It's probably using

You must be registered to see the links

; in which case I'm not sure if it could be used for our purposes, at least not easily. But it's a super interesting feature nontheless.



I can at least picture it used to contort the vagina/anus based on penetration, not as pretty as morphs but easier and more versatile. could be used to deform flesh on impact to augment the flesh ripple effect, could be used to increase versatility of character creation tools possibly even replacing many morphs, you could use this to create foreskin quite easily too, bulging of the stomach and neck based on penetration... I'm sure i'm only scratching the surface, there is heaps of potential in this. I wonder if it at least works with morphs... It should, because it works through control rig i think.

Edit: Hmm I wonder how expensive this is, it might be possible to create a fake jiggle effect with this by attaching deformers to the appropriate bones and make them trail behind the animation's movements. Although by that point we'd just be imitating bone based jiggle.

 

[razfaz]

Yeah, it is frustrating to see only the results without sauces. I feel the same...

Therefore, here is some food (attached PDF) which can make Your coding hands and brainwork dirty
I was testing the new Grok3 Release and remembered this "SoftBody" task and used some tokens on it for research and suggestions.

Below (attached PDF) is a detailed, full example for an optimized SkeletalSoftBody project in Unreal Engine 5.5.x as a result of this convo.

[README]
This example implements a performant real-time softbody physics solution for a Skeletal Mesh using a custom Position-Based Dynamics (PBD) solver in C++, avoiding Blueprints and Chaos Flesh in Unreal.

I have avoided Blueprints because of full control and possible improvements. Later on it can be converted to a custom Blueprint Component anyway.

I’ll provide all necessary code, asset creation steps, and setup instructions to get a wobbly creature limb and bouncing gelatinous blob running in Unreal using this approach. The focus is on performance while maintaining simulation quality.

Deformation will be solved via interpolation which is implemented as a custom Material (HLSL Vertex Shader) for performance reasons.

Hav fun guys and happy creative coding.

 

[Velomous]

Yeah, it is frustrating to see only the results without sauces. I feel the same...

Therefore, here is some food (attached PDF) which can make Your coding hands and brainwork dirty
I was testing the new Grok3 Release and remembered this "SoftBody" task and used some tokens on it for research and suggestions.

That's interesting, but does it work? My experience with ai as code assistants is that they're extremely unreliable for it. You can sometimes get it to spit out working demo code, like say pong written in python, but something as specific as this, softbody skeletal mesh in unreal engine speicfically, especially something that hasn't been done a thousand times before and doesn't have working examples online for the ai to draw on; I will be truly impressed if you managed to get working code out of it for a task as esoteric as this.

Using c++ would indeed be the only way to do something like that though, blueprint is great but only works within constraints already created for it in c++, so c++ is the only way to expand blueprints.

 

[JigglySquish]

Well guys, looks like the holy grail might have been found for us.


Of course, like always, the video says virtually nothing about how it's done. Frustrating shit.

This is what it says:


It's probably using

You must be registered to see the links

; in which case I'm not sure if it could be used for our purposes, at least not easily. But it's a super interesting feature nontheless.

View attachment 4500028

I can at least picture it used to contort the vagina/anus based on penetration, not as pretty as morphs but easier and more versatile. could be used to deform flesh on impact to augment the flesh ripple effect, could be used to increase versatility of character creation tools possibly even replacing many morphs, you could use this to create foreskin quite easily too, bulging of the stomach and neck based on penetration... I'm sure i'm only scratching the surface, there is heaps of potential in this. I wonder if it at least works with morphs... It should, because it works through control rig i think.

Edit: Hmm I wonder how expensive this is, it might be possible to create a fake jiggle effect with this by attaching deformers to the appropriate bones and make them trail behind the animation's movements. Although by that point we'd just be imitating bone based jiggle.

We've been able to use the deformer graph for a while in Unreal, what 5.5 is giving us is the ability to dynamically add deformers into Sequences.

So, before, you'd need to create your deformer graph and then add it to your character BP, now you can create a deformer graph and add it to a specific sequence as a track. This means that if you need a shot specific deformation of some kind - you can add one fairly easily. It's a tool for cinematics, not gameplay animations.

Deformers graphs are not really like a blueprint graph, where most of the code is in the graph language. They tend to just be a bunch of variables linked to an hlsl script node. That's good, because HLSL is much faster than a graph language. The graph is just connecting things together.

This IS a really cool thing for making smutty animations in Unreal, and I am planning on exploring this a lot more in the next little while.

 

[razfaz]

[Regarding SoftBodyDemo.PDF]

You all are lazy, it's all there written down!

*sigh*, apparently I|you have to do everything my|yourself...

In the meantime I have made some improvements regarding Shaders, ported the code into an "unreal plugIn" and was able to push more workload to the GPU.

I'm in the middle of coding and need to correct some booboos aka having some nice debugging sessions. Apparently Grok (evolution 3) is still stuck with Unreal 4.x... But no problem, that shows me that we Coders won't become obsolete so quickly. We'll see how this new adventure will end.

Q: What did AI for me?
A: It showed me new, alternative ways to tackle a problem! It is inspirational and that's really cool!

PS: Again... you guys are lazy bastards!

PS2: The topic of shaders is my Achilles heel. So, Is there anyone here who is a Crack in shaders?
Or do you know someone who is?

Because in my opinion, outsourcing the computing work to the GPU is the only sensible way to achieve performance + quality.
Unfortunately, I am unable to check the AI generated Shader code because of lack of exp. in this spec. area. ;(

---
Attach:
- Latest xAI Prompt regarding this Topic:
"This implementation provides a fully functional, GPU-accelerated softbody physics solution for Unreal Engine 5.5, optimized for human skeletal meshes while maintaining simulation quality. Let me know if you'd like to add specific features like collision handling or volume preservation!"

 

[Microtom]

Attach:
- Latest xAI Prompt regarding this Topic:

You might have more success with the new Claude

 

[razfaz]

Personally I do use VSCode + Roo(Cline) + someMCP's + incept5/Claude (locally)

Roo = (

You must be registered to see the links

) <-- Who may be interested

How ever, I just wanted to share my grok3 online exp. on this spec. topic.

PS: My current Model running locally atm for the interested...:



PS: I fear we get aside the main topic, /my bad
Never the less it can help to solve the problem!

Looking forward to Claude 3.5 or 3.7 (whateverr)

 

[darkevilhum]

I took the plunge and decided to try one of the earlier mentioned approaches to character physics. Not quite a soft body solution, but the results are performant and very good looking. (Will update with visual references when I can) but TLR, if you want a viable, performant and fairly easy to implement method, I would suggest you try multi-bone physics per "jiggle" zone, as used in various mod communities for mod-able game like Skyrim.

A simple example being, rather than having a single bone for say a breast which is then given some constraint based physics and a collider, you rig/skin a small chain of bones. I did this with three, breast1,breast2,breast3. These have an influence which decreases as you go inwards from the nipple and the results with a decently setup physics asset are really good.

 

[razfaz]

I took the plunge and decided to try one of the earlier mentioned approaches to character physics. Not quite a soft body solution, but the results are performant and very good looking. (Will update with visual references when I can) but TLR, if you want a viable, performant and fairly easy to implement method, I would suggest you try multi-bone physics per "jiggle" zone, as used in various mod communities for mod-able game like Skyrim.

A simple example being, rather than having a single bone for say a breast which is then given some constraint based physics and a collider, you rig/skin a small chain of bones. I did this with three, breast1,breast2,breast3. These have an influence which decreases as you go inwards from the nipple and the results with a decently setup physics asset are really good.

We are all on an journey and I think your curiosity is great! The "Jiggle Task" reminds me of the good old Skyrim times incl. custom skeletons etc.
and Your multi-bone physics setup was the total win back then - it was performant and more or less good-looking.

That was the best practice back then, however, we now have more computing power (GPU) at our hands and the ambitions are increasing!

As You can see You’re already nailing an aspect and tricky part of game dev, and honestly, this is the kind of thing that hooks you into going deeper and what we are talking about in this thread.

The past best practices are not good enough no more and thats why we are all here talkin' and sharin' some snips of wisdom.
Unreal's available Chaos API is very good, but not good enough for dynamic (ev. animated) human SkeletonMeshes.

The Unreal Niagara Particle Simulation can be an alternative way to tackle this task, but sadly I have'nt had much time to dive into it.
As far I can tell (after some superficial research), this could be a promising and verry good approach! (Note to myself: "Check this out and dive deeprrr!")


PS_1: I'm all in for throwing all the heavy computing for realtime "SkeletonMesh deforming" to our GPU's with the help Shaders, but that's me and my latest approach atm.

PS_2: I'm bit old, but still have the will of learning another high level lang, in this case for Shaders (HLSL), only because of this Thread, damn it!
Just for fun.

 

[animetiddyenthusiast]

Ah I see, there is no collision in that video. It's purely a shader effect being triggered by either a world position or a trace.

The only way to get an effect like that in UE5 with real collision would be to actually use their Chaos Flesh (which isn't really game or even runtime ready) or to create your own fully fledged true soft body system with collision which the NS dev definitely didn't do. He said as much himself and the work involved for that is quite a lot. You'd effectively need to recreate your own mini physics engine either in shader code or in c++ which then drives your shader soft body system.

With the stuff I've figured out so far, I can remake the effect in that video very easily, it's just not high on my priority list atm. As I'm not focusing on stuff that is more VR inclined. I'm trying to make a more general jiggle/soft flesh system atm where the skin will react to anything specified.

Hold up why is this what we're aiming for when Vam still has better physics than any game in unreal? Focus on beating Vam first

 

[animetiddyenthusiast]

[Regarding SoftBodyDemo.PDF]

You all are lazy, it's all there written down!

*sigh*, apparently I|you have to do everything my|yourself...

In the meantime I have made some improvements regarding Shaders, ported the code into an "unreal plugIn" and was able to push more workload to the GPU.
...

PS2: The topic of shaders is my Achilles heel. So, Is there anyone here who is a Crack in shaders?
Or do you know someone who is?

Because in my opinion, outsourcing the computing work to the GPU is the only sensible way to achieve performance + quality.
Unfortunately, I am unable to check the AI generated Shader code because of lack of exp. in this spec. area. ;(

---
Attach:
- Latest xAI Prompt regarding this Topic:
"This implementation provides a fully functional, GPU-accelerated softbody physics solution for Unreal Engine 5.5, optimized for human skeletal meshes while maintaining simulation quality. Let me know if you'd like to add specific features like collision handling or volume preservation!"

Ok hang on a sec I did not realize how far this thread has come. I may be your shader man. I'm more about c++ and d3d than hlsl and shadergraphs but I can completely reimplement optimal hlsl and I can read generated or compile code and somewhat make sense of it.

Recently been learning compute shaders for skinning and morph targets and been diving into the actual unreal source code quite a bit to make sense of how they do it. If you can share your project I can take a look. I see you shared lots of resources which are nice and but I keep myself pretty busy so if you want potential help it would be best if you sent a project ready to go. I've been mainly interacting with unreal sdk with code by hacking released games, so im a bit slow with the editor right now

 

[razfaz]

Ok hang on a sec I did not realize how far this thread has come. I may be your shader man. I'm more about c++ and d3d than hlsl and shadergraphs but I can completely reimplement optimal hlsl and I can read generated or compile code and somewhat make sense of it.

Recently been learning compute shaders for skinning and morph targets and been diving into the actual unreal source code quite a bit to make sense of how they do it. If you can share your project I can take a look. I see you shared lots of resources which are nice and but I keep myself pretty busy so if you want potential help it would be best if you sent a project ready to go. I've been mainly interacting with unreal sdk with code by hacking released games, so im a bit slow with the editor right now

Yeah, will share ofc.
A lot have changed in the API since then and the plugin does not Build using the latest API.
Currently I'm migrating the whole Thing to 5.5.3.
Can take some time, because my freetime is sadly also very limited.

 

[razfaz]

Ok hang on a sec I did not realize how far this thread has come. I may be your shader man. I'm more about c++ and d3d than hlsl and shadergraphs but I can completely reimplement optimal hlsl and I can read generated or compile code and somewhat make sense of it.

Recently been learning compute shaders for skinning and morph targets and been diving into the actual unreal source code quite a bit to make sense of how they do it. If you can share your project I can take a look. I see you shared lots of resources which are nice and but I keep myself pretty busy so if you want potential help it would be best if you sent a project ready to go. I've been mainly interacting with unreal sdk with code by hacking released games, so im a bit slow with the editor right now

Had some freetime today and felt the need for some AI-Assisted-Coding.

This challenge is getting a PITA because of the need of performance.

The more I try to put calcs to the GPU for the performance sake, the more I have to struggle with lowlevel things like parallel Shader-Buffers and its management, Memory Addressing and so on. Which I expected, but have to rethink it partly now.

The more I get LowLevel the more it will make such a plugin very hard to maintain in the future.

So, I decided to revert some performant LowLevel Code in exchange of performance, because I don't want to get back everytime the UE API changes it's version just on the 3rd digit (x.x.1)

So, thats my current state and journey for today|night.
Grok3 helped me a lot porting the code from UE 4.X to 5.5.X and spared me ages of time fixing things just by discussing debugger and|or exceptions output.

Again, I can recommend! 10x timesaver, at least for the ones who are more or less fluent in UE-API & C++ - which I'm not, but have always the learning mode ON - and therefore able to fix and maintain such mixed code of Human+AI Code.

God, I fear Oldschool-Coders will have a hell of work todo in the near future to fix all the AI generated code produced by Ppl. who can't maintain and fix the created code created and may be used by consumers or companies by a larger scale.

How ever, here the "BASIC" Shader (HLSL) that I use for testing (see attach). The Shader is very Basic, because I'm focused on the C++ Implementation of the Plugin first, which still throws runtime memory acc. vios at the moment of writing... *argh* -.-

 

[darkevilhum]

We are all on an journey and I think your curiosity is great! The "Jiggle Task" reminds me of the good old Skyrim times incl. custom skeletons etc.
and Your multi-bone physics setup was the total win back then - it was performant and more or less good-looking.

That was the best practice back then, however, we now have more computing power (GPU) at our hands and the ambitions are increasing!

As You can see You’re already nailing an aspect and tricky part of game dev, and honestly, this is the kind of thing that hooks you into going deeper and what we are talking about in this thread.

The past best practices are not good enough no more and thats why we are all here talkin' and sharin' some snips of wisdom.
Unreal's available Chaos API is very good, but not good enough for dynamic (ev. animated) human SkeletonMeshes.

The Unreal Niagara Particle Simulation can be an alternative way to tackle this task, but sadly I have'nt had much time to dive into it.
As far I can tell (after some superficial research), this could be a promising and verry good approach! (Note to myself: "Check this out and dive deeprrr!")


PS_1: I'm all in for throwing all the heavy computing for realtime "SkeletonMesh deforming" to our GPU's with the help Shaders, but that's me and my latest approach atm.

PS_2: I'm bit old, but still have the will of learning another high level lang, in this case for Shaders (HLSL), only because of this Thread, damn it!
Just for fun.

Given that I've now tried an implementation of all the methods discussed in this thread including a shader based approach with a custom physics solution (not the shader that I shared earlier) I think my personal choice to fall back on multi-bone physics was at least worth sharing. It just ends up being the most realistically usable solution for any project and without any scalability concerns.

The only circumstance I think where it's still worth chasing a soft body solution would be in a situation where the project only has 2-3 max characters running the soft body simulation at a time. Ontop of this, you'd likely want some very close camera angles or some specific game feature/mechanic to take advantage of the soft body, such as some kind of interaction or touch. VR is the best example of this. For anything else, I wouldn't recommend devoting development time/resources to this.

The nice thing about this solution is that it meshes perfectly with all of the existing tooling in UE5. So with some decently setup physics shapes in the physics asset, all of these "soft" body parts collide with everything else, including other actors.

Examples of my results (3 bones per breast, 3 bones per glute, 2 bones for the thighs (front and back))
The difference in jiggle is subtle at these constraint values (I'm going for something a little more realistic/subtle for this particular character) however the difference is night and day between a single bone constraint and this.

Spoiler: NSFW

You don't have permission to view the spoiler content. Log in or register now.

 

[razfaz]

@darkevilhum
Aye! From my personal view this looks very, very good! Good Job my friend!

Just silently asking:
- What about Actor/SkMesh: Collision, Morphs, Animations?

 

[darkevilhum]

@darkevilhum
Aye! From my personal view this looks very, very good! Good Job my friend!

Just silently asking:
- What about Actor/SkMesh: Collision, Morphs, Animations?

You mean with relation to the multi-bone technique? Yep those all work fine too.

The only caveat I've encountered is if you use morphs for very significant character customization, such as normal breasts to huge breasts. In this circumstance, the collision shapes for the breasts would need to be updated to match the new size.
This isn't too difficult to do. But can become a cumbersome bit of work if you have a lot of these.

But generally speaking, using morphs for customization in this way is easily offset by instead using bone scaling (which is the OG/Standard way of doing character customization for the body).

In my current project I'm using a mix of bone scaling and morphs for character customization because morphs make more sense for some things like finer details/muscle toning etc, and less for others such as a general size/width/depth/length.