MakeHuman NextGen Basemesh Talk
Posted: Sun Apr 21, 2019 5:41 pm
As I've been working with MH, Blender and UE4 for some time now, I've recognized more and more, where the flaws of the current default mesh are.
I've to admit that the bodyshapes of MH-Characters have the best realism/naturalness of all character creation tools available. We should keep that!
But the current base mesh topology/resolution (default rig with adult female genitalia topology, preseve volume/Dual Quaternion Skinning) has some weak points, if you want to make different poses/animations (yoga,stretching arms) for game development.
For example bending knees to extreme positions or raising arms doesn't look as good as it could be.
I've seen some tricks mentioned in the forum that work inside of blender like having multiple armatures for one character and some other blender tricks/modifiers (smoothen,subsurf), but it's for rendering static scenes and blender only.
Especially the knee and shoulder area don't deform well.
And while doing base mesh optimizations, we should also take Dual Quaternion Skinning into consideration, because it's more advanced than Linear Blend Skinning.
Dual Quaternion Skinning can be used with Unreal Engine: https://www.daz3d.com/forums/discussion ... -inside/p4 ,
as far as I know even with Unity 3d: https://forum.unity.com/threads/dual-qu ... ty.501245/
and it's the standard skinning method in CryEngine. And it works with blender as well: "preserve volume". DAZ utilises it too.
It seems as if the characters need a higher resolution, mainly in the knee, butt and shoulder/armpit/breast/clavicle area.
Simply applying targets won't help here unlike with other body parts, where the resolution is sufficient for adjustments by using target sliders.
Even making a shape key has it's limits with the low mesh resolution. And normally shape keys should be used only as a last resort for small corrections, where the rig has its limitations, because they are problematic, esp. with multilayered clothing.
At least the knees and legs can be reweighted to look more or less smooth.
Applying the subsurf modifier in blender is overkill, because the mesh resolution will be unnecessarily high in most mesh parts and for
most (even higher end) computers it would be overkill having a tricount of >135.000 (without clothes) per character in an open world scene with many other characters.
And it would be still not optimized.
Knee area problems could be theoretically overcome by adding additional edge loops (or a more intelligent knee topology), but for the shoulder it's more difficult.
In my opinion most weak points can be solved by increasing the mesh resolution in the critical areas+more intelligent topology+better bone weighting.
A good mesh for comparison of topologies is the DAZ Genesis 8 Mesh, which is optimized for Dual Quaternion Skinning.
Personally I don't like the daz body meshes, because they are too stylized/doll like and don't look like real human bodies, but we can nevertheless
study the rig, topology and weighting approach.
If you want to compare it with makehuman in blender, then simply load a gen 8 default character in daz studio, export it as .dae and import into blender. (it will come without shape keys, but that's ok for comparison purposes)
Now the question is:
Would it be possible to make an additional experimental base mesh, which doesn't even need to fit the current clothing assets if that would be currently too problematic?
I've heard that new topologies can be made, but only without additional vertrices or by subdividing the whole mesh (overkill)?
What do you think about it? How could we refine the current base mesh in order to look more clean and up to date?
Image 1: As you see the daz knee is superior to the mh knee, but reweighting can "heal" the mh knee to a certain degree (and even the leg looks now ok)
Image 2: MH Character has a horrible bulge when raising arms, that could be overcome by a shape key, but the mesh is still not optimized and lacks resolution (looks bad, esp. in game engines (shadow artifacts etc.))
Image 3: Comparison of MH and DAZ Mesh. See the knee differences and look at the clavicle,shoulder,arm and armpit!
Goal: A Default Mesh with optimizations in knee/shoulder area (mesh resolution, mesh topology, weights...) similar to DAZ but with keeping the current MakeHuman Default Body Shape!
I've to admit that the bodyshapes of MH-Characters have the best realism/naturalness of all character creation tools available. We should keep that!
But the current base mesh topology/resolution (default rig with adult female genitalia topology, preseve volume/Dual Quaternion Skinning) has some weak points, if you want to make different poses/animations (yoga,stretching arms) for game development.
For example bending knees to extreme positions or raising arms doesn't look as good as it could be.
I've seen some tricks mentioned in the forum that work inside of blender like having multiple armatures for one character and some other blender tricks/modifiers (smoothen,subsurf), but it's for rendering static scenes and blender only.
Especially the knee and shoulder area don't deform well.
And while doing base mesh optimizations, we should also take Dual Quaternion Skinning into consideration, because it's more advanced than Linear Blend Skinning.
Dual Quaternion Skinning can be used with Unreal Engine: https://www.daz3d.com/forums/discussion ... -inside/p4 ,
as far as I know even with Unity 3d: https://forum.unity.com/threads/dual-qu ... ty.501245/
and it's the standard skinning method in CryEngine. And it works with blender as well: "preserve volume". DAZ utilises it too.
It seems as if the characters need a higher resolution, mainly in the knee, butt and shoulder/armpit/breast/clavicle area.
Simply applying targets won't help here unlike with other body parts, where the resolution is sufficient for adjustments by using target sliders.
Even making a shape key has it's limits with the low mesh resolution. And normally shape keys should be used only as a last resort for small corrections, where the rig has its limitations, because they are problematic, esp. with multilayered clothing.
At least the knees and legs can be reweighted to look more or less smooth.
Applying the subsurf modifier in blender is overkill, because the mesh resolution will be unnecessarily high in most mesh parts and for
most (even higher end) computers it would be overkill having a tricount of >135.000 (without clothes) per character in an open world scene with many other characters.
And it would be still not optimized.
Knee area problems could be theoretically overcome by adding additional edge loops (or a more intelligent knee topology), but for the shoulder it's more difficult.
In my opinion most weak points can be solved by increasing the mesh resolution in the critical areas+more intelligent topology+better bone weighting.
A good mesh for comparison of topologies is the DAZ Genesis 8 Mesh, which is optimized for Dual Quaternion Skinning.
Personally I don't like the daz body meshes, because they are too stylized/doll like and don't look like real human bodies, but we can nevertheless
study the rig, topology and weighting approach.
If you want to compare it with makehuman in blender, then simply load a gen 8 default character in daz studio, export it as .dae and import into blender. (it will come without shape keys, but that's ok for comparison purposes)
Now the question is:
Would it be possible to make an additional experimental base mesh, which doesn't even need to fit the current clothing assets if that would be currently too problematic?
I've heard that new topologies can be made, but only without additional vertrices or by subdividing the whole mesh (overkill)?
What do you think about it? How could we refine the current base mesh in order to look more clean and up to date?
Image 1: As you see the daz knee is superior to the mh knee, but reweighting can "heal" the mh knee to a certain degree (and even the leg looks now ok)
Image 2: MH Character has a horrible bulge when raising arms, that could be overcome by a shape key, but the mesh is still not optimized and lacks resolution (looks bad, esp. in game engines (shadow artifacts etc.))
Image 3: Comparison of MH and DAZ Mesh. See the knee differences and look at the clavicle,shoulder,arm and armpit!
Goal: A Default Mesh with optimizations in knee/shoulder area (mesh resolution, mesh topology, weights...) similar to DAZ but with keeping the current MakeHuman Default Body Shape!