I am currently working on a muscle simulation workflow using Chaos Flesh in Unreal Engine.
After running the muscle simulation, I generated a Geometry Cache and successfully trained an ML Deformer model. However, after applying the ML Deformer, the muscle deformation causes the mesh to penetrate/intersect through the Skeletal Mesh clothing that previously fitted the character perfectly.
In earlier versions like UE 5.6, the standard workflow was simulating the cloth in external DCC tools (like Marvelous Designer), exporting the result as USD/Geometry Cache, and training the ML Deformer with it.
However, according to the UE 5.8 release notes for Chaos Flesh and Chaos Cloth, it mentions:
“Support for procedural interaction between layers (Skin, Muscle, Clothing)”
“Integration of Cloth and Flesh solvers into a single deformable core”
Based on this, I assumed that UE 5.8 allows an integrated workflow where muscle, skin, and cloth layers can interact procedurally directly inside Unreal Engine without relying on external DCC tools for cloth simulation caches.
However, I couldn’t find specific Dataflow nodes in Chaos Flesh related to Chaos Cloth interaction, nor could I locate any Chaos Flesh nodes within Chaos Cloth.
Could anyone explain or guide me on:
How to specifically set up this layer interaction (Skin, Muscle, Clothing) in UE 5.8?
Do we need to link them via specific Dataflow nodes, or is there a new Asset/Component setup required for this procedural collision?
Any documentation, node setups, or advice on this workflow would be greatly appreciated!
Yes it is expected that the cloth will go through
The cloth collides with the physics asset and this is resolved on the CPU side
the ML deformer runs on the CPU / GPU side ( on the CPU they output morph target curve values )
Even if the collision is using the actual skeletal mesh itself ( we have such an option for cloth ) it does not account for morph target yet and so the cloth may go through any geometry that is far above the pre-deformer skeletal mesh topology
Depending on your cloth setup, you provide a mask to the ML Deformer for the regions that are under the cloth ( that would require painting a map on the skeletal mesh for that purpose if I recall well )
But that would mean that you would not see deformation underneath the cloth
Another idea is using collision thickness to potentially account for the maximum deformation you may get from the deformer, but that may result in unrealistic cloth floating above any undeformed skin
Regarding the releases note, thank you for pointing out those two points
That is a mistake on our end, the unified solver is in active development but did not made it to 5.8 because of other priorities
So unfortunately, in 5.8, it is not yet possible to to make a cloth type skin to interact with the fat layer or muscles
This will be available in later release, sorry again for the miscommunication
We have informed the documentation team, and the note will be updated accordingly
Thank you so much for the clear and detailed response, as well as clarifying the status of the unified solver in 5.8! It helped me save a lot of time trying to find non-existent node connections.
I have a few follow-up questions regarding the current 5.8 capabilities and standard workflows:
Timeline for Unified Solver & Layer Interaction: Is there a tentative timeline or target version (e.g., 5.9 or UE 6) for when the unified solver and procedural interaction between skin/fat/cloth layers might be introduced?
New Features in UE 5.8 Chaos Flesh: Are there any newly added Dataflow nodes, performance optimizations, or features specific to Chaos Flesh in 5.8 that we should take advantage of for muscle/fat simulations?
Best Workflow for Head and Hands with GenerateSurfaceBinding: In tutorials, I saw the skin being bound to the muscle/fat Collection using GenerateSurfaceBinding. However, areas like the head and hands do not have underlying muscles—they are essentially just skin over bones. As a result, those vertices fail to bind properly to the simulation Collection, causing severe stretching or floating artifacts.
Currently, my workaround is splitting the character into two Skeletal Meshes: I simulate and train the ML Deformer only for the body (torso and limbs), and then combine it with the intact head/hands mesh at runtime using Opacity/Color Masking.
What is the recommended/standard workflow for this in Unreal Engine?
Is there a way to bind the full Skeletal Mesh (including head and hands) to the simulation in a single Dataflow setup (e.g., by falling back to bone weights for non-simulated areas)?
Or is splitting the character into two Skeletal Meshes and combining them with masking actually the intended approach when using ML Deformer?