Recruiting for Makehumans Future!

[RAS_MH]

Fixed.

Good work! Checked it out, it's all starting up now. Thanks for the quick fix.

[WojackOWL]

This work is incredible! I just made a Github Issue asking what was going to be different between this and the current MakeHuman version. I am so excited of what's going on. Is there going to be custom shader and tone mapping support?

[punkduck]

This work is incredible! I just made a Github Issue asking what was going to be different between this and the current MakeHuman version. I am so excited of what's going on. Is there going to be custom shader and tone mapping support?

That's not an issue ... that is my lack of information maybe

Phew, I struggle hard with that openGL stuff. Especially the idea of using an in-between geometry shader for normals was a rather stupid decision, maybe I can do that on the vertex normals (the normals calculated in the point met from the surrounding faces or I must redo that completely). The problem is, now I got all faces flat but with the correct normals on it

I need to emphasize, that most of our assets are rather "cheap", so no metal-map etc. I added those already for a few of my assets at home to at least test it. For that "wonder-woman character", since normal clothes usually are not made from metal . The exporters now should already use it, inside makehuman it is not as easy. If one combines assets with phong, matcap (litsphere) and pbr shader (the one from the openGL tutorial), then one has a lot of different light situations. For my Ava character I also added light, but in a game they use an overlay to do bloom effects. And until now I have no self-shading. Metal reflects the skybox, that looks okay, shiny skin creates funny effects ... so I still avoid that. One can use colored lamps to get a different tone, but then all assets change the tone. The big shader engines (like the editor in Unreal Engine) I can't realize, the problem would be, like in Blender, that hundreds of effects will look fine in that program, but cannot exported so it will fit into another system.

At the moment I try animation corrections but these do not yet work correctly, I have still problems with the axes. There are other bugs, some I am aware of. And bugs occur here and there, when versions change or with a different OS or maybe graphics card.

[tomcat]

the problem would be, like in Blender, that hundreds of effects will look fine in that program, but cannot exported so it will fit into another system.

These are issues of the project paradigm — goal setting. It is usually formulated before the work begins. What is the purpose of the new MakeHuman:

• It is a self-sufficient program

or

• is intended to create a billet, a dummy that will be finalized in another program (Blender, game engine)?

In the first case, it should be possible to create an environment. This is already actually something from the analogs of Blender (or similar 3D CCP\game engines).

In the second case, it is important to observe the possibility of transferring the result to the target program.

[punkduck]

Hi

Well I know my own goals after 2 years. Lets start with a very simple example. Without MakeHuman I would never have started Blender or Unreal Engine.

And I know a few people who need an easy approach first. When one starts Blender, then there is a cube. I deleted that crap from disk in the beginning. Years later I tried again. And it took some time. The game-engine? Same. There is that mannekin in Unreal, I do not want to know how many people download it, try with that sad guy and give up after some time and then delete it. So there should be some "out of the box stuff" available. The more sophisticated the hobby becomes, the more applications I need.

I will not go on with that discussion if s.o. considers the program as self-sufficient or just as a character morphing system (which it mostly is). For a morphing system a simple litsphere shader + albedo/diffuse texture would be enough. Also that orthogonal perspective we had in last version. A few poses to test etc maybe. But it would look like crap.

So a few things had been desperately needed, like perspective. Lets start with this: I tried to copy certain real persons. From photos of course. They mostly looked good in Makehuman first, but when in Blender + perspective they all had big round heads.

I know it always looked totally different, when in target-system. Diffuse textures looked blotchy, wet-map produced funny effects (wet on black + a lamp = white) ... In the end I needed a new texture in the target-system. Blender cycles worked fine, because the ray-tracing concept works totally different. But neither Godot nor Unreal Engine created sth similar. So it was interesting to add a shader similar in makehuman (game engines work with GL-Shaders as well). The target-system can correct shaders with additional elements (like filters) of course. But the more elements, the worse the runtime.

So at least a few more featutes of modern shaders shall help. Before I can export stuff to another system, I need to be sure that I do not have to repeat this a dozen times. So, best one can test it before export. Meanwhile some of my assets had the metal-roughness textures and normal-maps needed for glTF and meanwhile I am also able to see them in MakeHuman as well. Animations I did with one character. Now the hands of another character intersect with body, but it can be corrected and now exported as a walk for the other one. One should be able to save it in MakeHuman via glTF for Blender or a game engine (need to do that still, until now without correction),

Especially this took a few weeks to solve that problem, but of course, math is easy, once I understood the principle. All vertical bones (so parallel to bone axes "y") worked without any correction.
Well the bone axes and the "y" axes are the same direction. The rest-matrix explains the orientation of each bone.

When I rotate the hand a bit, I would call that a "correction matrix". So the result was:

Multiply inverse rest matrix of a bone with correction matrix, then multiply with rest matrix of same bone ... this resulting matrix then is used in all frames of the animation like an offset (or overlay).
Until now I checked exports only in blender. This I will do with game engines as well (only glTF).

Here is a test with a some wetness (using a Metal-roughness map) and a normalmap fpr the face:

roughness and normalmap

So a few things to do still, mostly corrections of errors. One of the weird things had been the .rgb order of loaded files of metal-roughness textures. b= metal, roughness should be g ... and I found it on "r" ,,, the reflection of the landscape on metal did the same ... green and red had been swapped.

Greeting punkduck

[tomcat]

Blender cycles worked fine, because the ray-tracing concept works totally different. But neither Godot nor Unreal Engine created sth similar.

CCP and Game Engines are completely different things and occupy different niches. Blender can afford to render (and trace) leisurely. And engines work in real time and don't have such luxury as extra free time. By the way, this is the reason why a game engine on Blender is not possible despite all efforts.

[WojackOWL]

Blender cycles worked fine, because the ray-tracing concept works totally different. But neither Godot nor Unreal Engine created sth similar.

CCP and Game Engines are completely different things and occupy different niches. Blender can afford to render (and trace) leisurely. And engines work in real time and don't have such luxury as extra free time. By the way, this is the reason why a game engine on Blender is not possible despite all efforts.

Yeah, I agree with this as well. Even Unity doesn't really need raytracing for a lot of its use cases. Punkduck, have you ever thought about looking into other character creation software like Adobe Fuse, DAZ3D or Unreal's Metahuman for research to see what elements you want to bring over or improve on for this?

[punkduck]

Well

A PBR shader is not the same as ray-tracing. It is just an approximation and it is used inside e.g. Unreal as well as in other systems as well (also Godot btw). And it does not matter if Vulkan or OpenGL, the commands of the graphics card are the limit. And by default most graphic cards do not support real ray-tracing (be aware that some people consider a reflection sphere for mirroring effects in e.g. Unreal as ray-tracing). Games engines usually use pre-compiled light situations. This of course often needs two scenes of the same if you want to create day and night scenarios. I checked that PBR effects for godot as well. Unfortunately most YouTube videos just play around with the values using a sphere and do not use a texture of e.g an armor consisting of leather and metal parts or sth, else. When I would supply clothes here for MakeHuman, it would be nice to see buttons made of metal. Especially when the gltf exporter is able to export it with a metal-roughness map. This was the reason to create at least some of these effects.

On my Windows system there is a Metahuman. These characters are looking extrremly real. But they have a few problems. First problem: I do not want my characters to have 1 giga byte each. Next export a Metahuman to Blender and enjoy. I tried to create a bicycle animation for a Metahuman, Of course then, you can only use a simplified skeleton. All these cool 500 bones keeping the character in correct form you can forget. But when you delete them you need a new weighting. So the wonderful effects to get real face movements or even correct behaviour of shouldes are gone. These characters were not made that you can change things or add own animations in a simple way. This I tried with Elv some time ago. Or try to animate that face. No I am not speaking of grimaces in front of the own smartphone, try it manually.

Mertahuman as well as Unreal and most of the other engines are usually written in C/C++. Although I would be able to program that language as well, because I am pretty old and I remember times when there was no python at all. But we then cannot re-use old code. We decided for python and without numpy it would work very slowly. And then we needed some kind of predefined gui. Unreal comes with its own and that windows manager for the editor sometimes does some weird effects with either Windows or Linux native one
Python simply is slower and uses more memory. There are even worse languages for that purpose btw.

Daz3D is best for pre-defined characters or for those who buy all the stuff. A friend of mine is bothering about that they want to sell stuff, they try to avoid these characters to be used like we use them. It is not that open. One of our former members is the author of the diffeomorphic exporter, which btw. is not even "official". So yes, research. The source code all these programs is of course not free. So one may "steal" the concept but not the code.

So anyway the corrections now can exported to Blender with socket, export file format and with gltf. Now a different skeleton is still missing for glTF export. And I guess then I need to test with not only Blender but also with godot and unreal. Unity I do not have.

As an example I now change my fashion pose inside makehuman that my character really touches her head, and also the sub-divsion algorithm (only used internal in a kind of renderer for those without Blender or other software) is now mostly working. One of the problems has been fake-seams generated by wavefront import (identical UV coordinates where no seam is). The weird thing with that Charles Loop algorithm is, that they all give the examples without textures ... and the real seams are a problem, since you always need one vertex on the other side of the seam

posing and subdivision

Furthermore an example of a custom folder of targets. Btw. these targets also work with old makehuman. But this way it looks more seamless and one also can add own icons + incr/decr targets also generate only one slider with a "middle position". Simply done by using correct name patterns. These own targets can also be compressed to binaries (like the internal ones) to load them very fast. The character animation of walking even work with these shoes. Unfortunately she walks a bit slow. These shoes have a lot of geometry

custom targets

Greetings
Punkduck

[tomcat]

First problem: I do not want my characters to have 1 giga byte each.

Why? What size is expected?

For different conditions, there are different needs. Some need a low-poly model, some need a very detailed model. In the “old” MH this was solved by having a “proxy”. It was a decent solution at the time, but it was palliative and temporary, transitional. Low-polygonal models did not have their own ecosystem and infrastructures — there were no specialized assets for them (hair, clothes). With MH rebuild this issue could be solved radically. Well, at least try to solve it and lay down ways to solve it in the future. But there had to be a development plan for that.

[punkduck]

Metahumans?

Well you can ask AI for that. "why do metahumans need so much space".
Not a 100% happy with that answer. Of course I am not an expert, but in my opinion the textures are not the reason. These things one can check easily in the filesystem.

Metahumans are supposed to work in Unreal. They come with a post-processing engine, in my opinion all the stuff behind is really using a lot of memory. And adding smaller LODs is possible, but adding them means even more memory.
Of course when character is far away, a lot of bones are not working, also IK might be switched off etc. etc. etc. But as I said: these characters are ALSO there.

So if anyone is really interested, the concept behind is really completely different:
https://dev.epicgames.com/community/learning/tutorials/2JVl/unreal-engine-metahuman-skeletal-mesh-optimization-guide

And this is the head-rig. Try to work manually in Blender

https://forums.unrealengine.com/t/metahuman-export-to-blender-problem/1625897

Check videos in YouTube on what could be done. Of course one can export those guys to Blender. But then try to get happy with the rest, which is still there. So MetaHuman is mostly dedicated to it's purpose.

Btw.: Female Metahumans always come with a woolen sports bra. Fantasy characters with a deep neckline (which is pretty common) will look a bit weird though If you want to change that: the normalmaps on the skin have to be corrected as well .. and a few other maps etc. So Metahumans are mostly bible-belt ready.

Meanwhile the corrections also work directly on a posed character, before I was only able to edit them in A-pose. So our famous "sitting pose" finally works without the hand intersecting the knee (okay I turned her head as well).

glTF works with different skeleton meanwhile and animation is translated from our skeleton to e.g. CMU skeleton. Of course these translations are partly not as good as the ones with own skeleton. Maybe I can improve that still, but there are mathematical limits.