MetaHuman Creator

[SalvoSoftware]

Damn, Epic Games really copied MakeHuman or is it just a coincidence?

It makes me kind of sad that MakeHuman will be overwhelmed by MetaHuman, but it is really amazing. I wonder how many people will mistake MetaHuman with MakeHuman and will download the second instead

[bogdan666]

As nice as MetaHuman looks, it probably hasn't even full body meshes and will be a cloud only application, which means that you are stuck with just a few hairstyles, body types, clothing assets etc.
Whenever they decide to take the service offline, it will evaporate.
From the first look you get the impression that it is more like an in-game character creator than a fully fletched human creation software.
The realistic skin effect is just some node setup with PBR textures, which is nothing special and can be applied to any character.
The hair technology is not MetaHuman specific and can be used for MakeHuman characters as well.

But it is undeniable that MakeHuman needs to become better. The base rig and body mesh need a massive topology update for allowing realistic posing and animating.
This also means the introduction of bone driven corrective shape keys for joint rotations and muscle flexions.
The UV layout needs a complete rework (better balance: more resolution, esp. at the areas with flexible skin like armpits,knees etc.)
Last but not least the face expressions need an optimization too.

[joepal]

I agree with pretty much all you say, bogdan. We'd need to recruit an artist who could put in serious man-hours to redo all the targets though. That is work that is likely to exceed months and approach years of effective work.

[fake_Person]

I agree with pretty much all you say, bogdan. We'd need to recruit an artist who could put in serious man-hours to redo all the targets though. That is work that is likely to exceed months and approach years of effective work.

id wish I could help, I would volunteer for free, but alas I know nothing

[JESTERRRRRR]

I haven't been able to find much footage of MetaHuman but like has already been said, it does seem to offer less control than MakeHuman. If i remember correctly one of the demonstration videos said you pick from 18 bodies, which is probably the area MakeHuman outshines it the most. I think it said you also pick from faces and blend them, but the video does look like you have some sort of handle system to edit faces. You pick hair and clothing, I suppose that is tied into the limited 18 bodies.

About the basemesh for MakeHuman. Assuming for a second a new base topology could be acquired, and assuming it is the same size/pose/shape (basically like it's shrink wrapped onto it), would it not be possible to transfer the targets? Pretty much like how a proxy is fitted, except we'd be saving the transformations of all the vertices in the new mesh for every target. I'd also subdivide the current basemesh before and after targets are applied to get a better result. The base skin texture could be baked over to a new unwrap.

The base rig and body mesh need a massive topology update for allowing realistic posing and animating.

I've seen a lot of human meshes recently. With regards to the topology (not the corrective shapes), could you go into a bit more detail? For example I took some quick screenshots of the base mesh and the MB lab one and just highlighted a few:



Are these the types of differences you mean, the different flow in areas that deform more, particularly the elbows for example? Also would you recommend a higher poly count?

Is it common practice for the head/face unwrap to share a texture with the rest of the body?

[bogdan666]

Sorry for not answering so long... (was playing too much with Unreal Engine and Daz3d)

A good example for clean and streamlined topologies is Daz3d. I would go even further and increase the mesh resolution of a daz3d character (base resolution Genesis 8 has around 35k tris) for the whole body by some 20-25 % (maximum) or at least at some focal points like elbows, knees, face and feet as one full subdivision would be too much. But at first we have to decide if we want to use linear or dual quaternion skinning. This will decide about how the bone driven shape keys will look like. Most important is that the topology density should be balanced and not look like improvised. (just adding some additional edge loops on a bad topology is a big no go)

Maybe the bodies need a rework as well in order to making them more average. (average natural people, maybe different bases for every (sub-)race (at least the faces). From this "average" base it will be easier to make different characters with deviating properties. The new body will need completely different body targets.... Maybe every sex needs a separate mesh topology (state of the art) And the nails have to be built into the body mesh.

For example Reallusion CC3 has released a nice addon with many new morph targets, it's called Ultimate Morphs. CC3 gives a good example on how characters need to be done. Unfortunately the characters don't deform very well, because they don't use bone driven morph targets/shapekeys for compensation of volume loss with bent joints and they have a somehow too cartooney look (esp. the hip area and the legs (they look like twisted in a way)) But at least they are optimized for linear skinning, in contrast to daz3d, which is optimized for dual quaternion skinning.

The skin uv layout needs a rework too. It's distribution is bad, because knees, butt, armpits etc. have a too low resolution, esp. with specific poses (bent joints). As first measure you would try to fit everything into one 4k texture and for additional details bake a 4 k normal map and later use a specular map etc. For pores etc. some additional repeatable texture can be used...

Regarding the ideal layout of the topology the MB lab one looks pretty good in comparison with the makehuman one and yes it's about the different flow in areas which bent more.
MBlab has definitively a too low resolution, because it works with some blender specific modifiers (not usable with real time game engines), so it would need a higher poly count, at least 35k tris * 1.2 or 1.25 (maximum), so the muscles (and shoulder blades, ankles, kneecaps, ears, elbows...) can be seen more clearly.
And we should not forget about the face / face expressions...

At the moment we needn't care about hair, clothing etc., because that's the easy part. Every hair can be imported and retargeted easily, so the first priority is really the body / body topology and a better rig, because the makehuman rig
has too many unnecessary bones instead of bone driven shape keys.

Maybe we should study the MetaHuman Faces/Bodies more (when there will be some bodies to look at) and see on what scientific knowledge they are based as it will probably become the most advanced and up to date character creation framework. (regarding the body topology and the body types).

Summary: New MakeHuman = Something like Daz3d, but slightly better overall mesh resolution. (for even smoother look and better visible anatomy/muscles)

[JESTERRRRRR]

No worries.

Thanks for the detailed reply, you have given me a lot to think about. I had not heard of dual quaternion skinning, I personally (I only use MakeHuman for game related stuff) would avoid it since the game engine I work with does not support it, neither do Unreal or Unity.

[bogdan666]

There is another character creation framework, which has been released not long ago (Human Generator). It's still not finished, but it would be worthy to study it's approaches on "edge flows" and base mesh body proportions/anatomy.
(I dislike the fact that the head has double vertices density of what the rest of the body has, because it "disrupts harmony")

At first I wouldn't even care about doing too many shape keys for body types, but instead getting two perfect body base meshes (one for each sex) + afterwards bone driven shape keys for body deformations + face rig
Later on different shape keys could be made for different basic body proportions, faces, ethnicities, ages etc.
As for the standard base pose, the A pose seems to give the best results (with linear skinning).

Link to "Human Generator"
https://blendermarket.com/products/humgen3d

The goal is to become better than Daz3d & Co., otherwise it will not be possible "to beat it". Currently out of all the character creation frameworks Daz3d has the most to offer, when it comes to creating and simulating a broad range of people. The base mesh is optimized for real time engines (LODs can be set automatically in UE4) and has now an even more advanced face rig (not checked yet). The bodies have a natural look and deform well, although the base mesh resolution seems a little bit too low in close up view (by some 20-25 %).

When it comes to doing (mocap) naturally looking advanced animations, realistic skins and body LOD / texture baking optimizations, then Reallusions CC3 has the best to offer, but personally I don't like the CC3 bodies, although the faces seem to look very convincing. An here again the body needs a better resolution, with exception of the head, which is perfect.

[JESTERRRRRR]

Thinking about this though, would it not be easier to stick with the base mesh just provide more optimal proxy meshes for export? That way very little would need changing. The current base mesh could be subdivided smoothly and all the targets recreated from the result if it needed to be more detailed, and you wouldn't need different targets for male/female (at least the ones that are currently not gender specific) - only different proxies. I suppose it does depend what level of control you are after but for the body I found MakeHuman gives me more than enough control (face has been a challenge). Would be cool to get some more races/ethnicities but considering each race has multiple targets for gender and 4 for age it's a lot of work.

[bogdan666]

The current mesh seems outdated. It's not only the topology, but the whole anatomy that needs a complete overhaul.
Base meshes with good anatomy are the foundation for good topologies, because the topology
needs to fit the anatomy. ("proxy" follows anatomy)

A simple subdivision won't do any wonders and will even "dilute" the base mesh further.

On what scientific foundation is the base mesh based? It has many anatomical problems like ankles, achilles tendon, toe shape,
knees, hips, ellbows, clavicle, armpits, shoulder blades, ears, eyes etc. and thus is irreparable.

Furhter it needs a basic muscle system for achieving basic "bone driven muscle flexion morph targets" (daz3d Gen 8 has it alongside the bone driven joint corrective morphs (technically it's both the same))
A base mesh needs a good anatomy, an advanced "aerodynamic" topology and a certain minimal "resolution" (polycount, slightly higher than Daz Base Mesh).
Daz3d is the best what we have right now, but it has it's problems as well in real time engines like a slightly too low resolution mesh, esp. related to knees or ellbows, when bent to extreme positions.

For a standard base mesh we need to define a "standard".
A standard (for realistic humans) can mean: "naturally looking, biologically healthy human of perfect/natural proportions".

Hard to explain, but you know what I mean: The characters should look like average people we meet everyday (the beautiful ones )or at least we met in the past (1990s etc.). aka "Next Door Girl".
You probably won't know what I mean (or know just from some old movie series), if you didn't grow up in Europe of early 1990s.

The standard base mesh for each sex is the foundation for all specific body types and deviations like different body sizes, aging, (sub-)races, proportions, deformities.... which are all achieved by morph targets.

An interesting example is the "Project Evolution" for poser which has nice looking base characters. (looks- and proportion wise (though a little bit outdated renderwise))
The mesh resolution seems to be even much higher (more like MetaHuman?) than that of the daz3d base mesh, but it is no problem if you use LODs in your game environment.

I've looked it up and according to the specs sheet it has a poly count of around 76k and an interesting advanced topology. The polycount is more "efficient" than the first subdivision of Daz3d, because here you get
more or less the same mesh details like with the first daz3d subdivision, which has a much higher polycount (136k tris) and thus is not usable for real time applications.

Link to Project Evolution Video: (more examples via google search)

https://www.youtube.com/watch?v=hkjbeDy ... genesisArt

Link to Renderosity page:

https://www.renderosity.com/rr/mod/bcs/ ... uct=127770

Maybe a cooperation with that given artist would be possible?