MakeHuman Female Portrait

Like many people, I am looking forward to the imminent update of MakeHuman 0.9.1RC1 http://www.makehuman.org/blog/index.php. I plan to use it with Blender to try and create a photorealistic female portrait; I have already started work on features that may be better modelled from scratch.

Hair was generated using Blender particles.
Eyebrows and eyelashes were constructed using Bezier curves and converting to mesh.
Eyes were constructed in five parts including a non-reflecting insert to avoid internal reflections; I omitted the lens and opted for a hole in the pupil.

The work to date has been assisted by means of the Blender manual, tutorials, this Forum and searching the net. At the moment I am struggling with a few things.

I am using four square area lights to simulate a window-shaped primary light in a studio setup. How do I get the area lights to reflect accurately on the cornea? Specular reflection only gives circular - not square - reflections. I’m vaguely aware of HDRI but don’t want to use Yaf(a)ray 0.1.0 because I believe it still doesn’t render Blender particles/hair. (I would use LuxRender or Indigo except for the same reason).

Is there a way to blur or randomize the edge of natural objects, say the join between iris and eyeball, to avoid a sharp boundary? I propose to use Gimp and do post processing if it can’t be done in Blender.

It’s a long way to go yet…

Attachments




After trialling YafaRay b299 and LuxRender v0.5 once again, I decided to stay with Blender internal renderer and try and make the most of its capabilities - among them the ability to render hair and particles. What also influenced me were the discussions that I read on the YafRay Developers’ Corner.

Reflectors have now been added to the studio area lights so that appropriate square reflections are given in the cornea. The blood vessels on the eyeballs have been made finer and less obvious and test renders are proving satisfactory so far.

Attachments


The proposed portrait studio setup is show in the image here. The studio is a box made from solid rectangles to avoid light leakage during raytracing (edit: it’s needed for Yafaray photon mapping but not for Blender Internal). Previously, the MakeHuman 0.9.1 RC1 *.Obj model imported on a relatively small scale but I don’t know yet what scale the updated MakeHuman models will be.

BU - 1 Blender unit scale
MH - MakeHuman imported model
Cf - face camera; Lens 75mm
Ct - for upper body; Lens 80mm
A1-4 - main area lights with emitting reflectors; Energy 0.04
A5 - fill-in area light with emitting reflector; Energy 0.06
S1-2 - hair highlight spots; Energy 2.0
S3 - background spot; Energy 1.0

Other developments on this portrait have involved SSS tests on a low poly version of an MH head; I’ve still not got the hang of SSS, nodes, scales and so on and I expect it will all come together once I combine SSS with appropriate textures. No images to post just yet - just dozens of renders of red ears!

Attachments


Oops. I thought that MH 1.0 was going to be released “… around Christmas (2008) sound good, at least for a <<technology preview>> release …” but recent information from the developers shows a much longer timescale. See http://makehuman.wiki.sourceforge.net/Release_RoadMap.

I will therefore work with 0.9.1 RC1 and see what I can do until MH 1.0 is issued.

A technical pre-alpha version of MakeHuman was released two days ago http://www.makehuman.org/blog/index.php. It currently has limited functionality, such as no poses and one of its features is a cleverly modified mesh. A rough comparison showed me that the old mesh would suit me just as well for a portrait. The scale remains the same and the .obj import still gives a relatively small model - hence the difficulty with SSS.

I watch the progress of the MH 1.0 project and wonder whether the guys are their wasting time developing a multitude of ethnic variants. I think they would be better, say, creating more advanced morphing tools for the few who want such figures or a large library of poses. Anyway, its their decision and we’ll have to wait.

So I’ll continue with 0.9.1 RC1 as last decided.

Attachments


A pose was worked out in MakeHuman 0.9.1 RC1 and the .obj file imported into Blender (only the Object, Group and Material buttons were enabled so that eyes, eyelashes and eyebrows could be deleted separately).

The mesh was modified before it was Subsurfed x2 then further mesh changes made. MH textures were modified in Gimp before use. No SSS has been done yet and it has to be combined with textures; getting the balance right is going to be interesting.

A lot more work to be done yet: hair, eyelashes, SSS, DOF, refine textures, refine bump maps, tweak lighting, refine mesh and add accessories.

I estimate the effort so far expended on this project is the order of 40 hours - 300 filesaves to date.

Attachments


…lol. Great job dude! I am also using make-human for some blender work . . . although I am quite scared of admiting to it. Keep up your work and look forward to progress! You certainly deserve top marks for the amount of work you put in . . . and it will look so good on your portfolio!

Work has proceeded on mesh modifications, skin texture, SSS, eyebrows and eyelashes. A check on the eyes show that progress is being made - but lots of work yet to do. The lower eyelashes need fine alignment and clumping, the upper eyelashes need to be thickened and lengthened as well as locating properly. The eyeball blood vessels seem to have got lost somewhere and need correcting. The conjunctive (what’s the correct word - the “third” eyelid) needs softening and making more reflective. The “tear” in the left corner needs attention and it is planned to introduce a wet interface between eyeball and lower lid.

Skin pores need reducing in diameter and depth. The skin texture and SSS balance may not yet be optimised so it will be kept under review. Eyebrows will probably be made slightly thicker and longer and less mature. Makeup will probably not be applied in order to keep an ordinary, young, fresh look.

Plenty of hours expended again - but isn’t it fun?

Attachments


try modeling your own female

The pose was modified and hands removed from view to avoid complications with additional skin and textile modelling and textures at this stage. Some modifications were carried out on the mesh to make some features asymmetrical - nostrils, lips and so on. Eyebrows were thickened in four or five sessions; same with eyelashes which are still being worked on and need more upward curves. New higher resolution (2048 x 1024) bump maps made separately for lips and eyes. Still having problems getting these correct when used with SSS. Current setting for nodes is 30/70 split between SSS and texture maps.

DOF is probably better using Gimp but it’s left switched on at present. Rendering times when hair is added are over 12 hours and the PC has crashed twice. Looking into compositing or baking or whatever may be needed.

Q6600 3GB RAM 2.4GHz 1TB HDD 256MB Video Vista 32bit

Attachments


More modifications to head and teeth meshes; eyebrows and eyelashes manipulated yet again. Seven textures are being used with SSS: bump, modified MakeHuman, skin blush, makeup, pores, lips bump and head gloss. A 70% texture 30% SSS balance seems about right at present. AO also in use; Sub preferred to Both. Eyebrows could do with still more work, eyelashes need turning up and clumping, teeth need the dark joints removing - SSS would probably work here. Automatic image save and automatic PC shutdown were discovered and now are used for the lengthy renders.

Big problems encountered when rendering head and hair where, typically, a render would crash after 12 hours. The head alone takes over two hours rendering on a quad processor. The hair consists of approximately 100k particles - so it adds greatly to the CPU load. I’ve used the CMD BCDEDIT setting for Vista to overcome the 1.5GB RAM limitation but it doesn’t appear to be effective. I don’t think that compositing will work in this portrait where the objects are in very close contact (assuming separate renders of head and hair) but I’ll continue to study the option. I’ll also think about abandoning the particles/hair approach and create a mesh-based hairstyle. If that’s the case I may as well utilise Yafaray or Luxblender which should be interesting. I don’t think either renderer supports Blender SSS - so it may be tricky.

Decision time.

Attachments


Devised a decision table of 8 popular renderers with 15 factors where Luxblender beat Blender Internal by 39 points to 38 - until I changed the ‘personal bias’ factor to make it a tie. Decided to stay with Blender Internal for a while longer.

Tried Macouno’s ReallyBigRender Python script http://www.alienhelpdesk.com as one method of sidestepping the Vista 32-bit 1.5GB limitation and it worked well with no crashes even though Blender showed 1540.80M (62.59M) in use at one time. Stayed with Blender particles/hair and finally managed to include most shanks of hair in the latest model and consequently this image took a total 24 hours to render; autosave and shutdown worked a treat. Post Gimp work done on DOF blur and lip gloss. Next step will include more hair, teeth fix (SSS doesn’t seem effective), lip gloss, more tests with SSS and textures. The model could be made to look more attractive - but that’s not the point - I want it to look realistic.

Attachments


Remodelled head and full figure in MakeHuman 0.9.1 RC1. Different approach to hair: part mesh and part particles. Dropped the use of SSS on skin for time being. 6 hour render in single frame, no crash. Gimp retouching. 600+ filesaves to date.

To do: further mesh asymmetry; more skin textures manipulation - especially avoiding plastic sheen and that strange hue; lip glossiness required (still haven’t full control of Spec map); better UV texture for hair on meshes; more particle hair mainly on hairline; eyeballs resizing or re-positioning; eyeball moisture line; reduce density on eyebrows; re-orientation of eyelashes; fiddle with lighting and so on … and so on …

Attachments


At last I think I’m getting away from the flat plastic look and more of a waxy appearance; at least it seems to be the better direction in which to obtain a more realistic skin shader. After much searching and studying of Blender, 3DSM, MentalRay, Maya and other sites I concluded that I inevitably have to work with some sort of change of hue and/or intensity in the skin shadow areas - other other words a sub-surface light scattering. After trying multi-texture shaders, multi-material nodes and various published SSS nodes I think this BDSkinNode which I’ve modified gives me the best image (can’t remember the originator of this one; possibly an early Pixelvore). Anyway, its a straightforward node setup that -even- I understand and therefore should be able to carry out further logical trials.

Also modified eyebrows, added hair, tweaked eylashes and sorted out lip glossiness control (unchecked UseAlpha and CalcAlpha in Map Image). Some renders were carried out with DOF but it always gave too grainy an image even at maximum settings so I’m using Gimp for post-op again. Much more to do on the hair - and skin of course.

Attachments


As usual the progress curve is an exponential one; equal amounts of effort appear to result in less and less improvement. Skin SSS scattering radii settings have been reduced several times with no apparent change in colour but I’ll continue to reduce the settings until a change becomes obvious. Now I’ve slightly better control over SSS its used on parts of the eye, teeth and pearls. (Typical: give a guy a new toy …). Decided to eliminate the pigtails and create an up-do/bun hairstyle where the number of particles can be minimized; a headscarf or baseball cap was considered at one point. More hairdressing still to be done. Earrings were added for a more feminine touch; ear piercing and earring refinements yet to be done as well as the many previously mentioned manipulations.

Attachments


Converting from Vista to Linux has delayed progress somewhat this last month. I have Ubuntu 9.04 64 bit running Blender 64 bit and benefit from the 1.3x increase in rendering speed - and Blender seems to run more smoothly (it just ‘feels’ better). Linux is being given a thorough two month trial before I convert completely - Ubuntu looks great so far.

Another benefit from the 64 bit OS is the fact that I can use all my 3GB RAM and am able to render a full head of hair. I saw the memory usage go over 2GB at one point and this image took 14 hours to render.

Further changes to the model include: bump map on teeth, more bump on lips, reduction in eyeball luminosity, improvement to eyebrows, asymmetry here and there and so on. In fact, there are many changes needed before my wife is anywhere near impressed.

Not finished yet.

Attachments


Hair, eyebrows and lashes are nice, the skin texturing is nice also. I think your material settings just need a little tweaking, perhaps a little bit more sss.

Also the standard make human teeth at the front are pretty awful, they look like those little minty chewing gums you can get. It’s not difficult to improve them, I would down scale them slightly, bring them closer together and move a few verts around in edit mode to make them more individual looking.

The only other thing I would say is that the groove in the eyeball socket where the cheek meets the bags under the eyes seems slightly too much to me. Overall though it’s looking pretty good, it’ll look great when it’s done.

A bit much info to read over now, but looks like a very methodical approach. I like that. :slight_smile:

http://blenderartists.org/forum/showthread.php?t=145904

If you haven’t seen that thread, you will certainly want to as it tests human skin in several renderers including Blender Internal. I’m currently working with skin a lot in my own thread too: http://blenderartists.org/forum/showthread.php?t=117247&page=18 and I plan to post some files in the next week or so hopefully so people can pick apart what I’ve done.

One thing to help you cut your render times down substantially - SSS blurs the lighting on the surface somewhat so you don’t need as many raytracing samples by a long way. Especially for a still because the shadows don’t need to be 100% consistent across multiple renders.

Something else that may help you is the SSGI node that is in development. As you are using 64bit, I can tell you that http://www.graphicall.org/builds/builds/showbuild.php?action=show&id=932 should be a build to look in to. Optimised as well, so it will render anything faster really.

All the best with the project!

I gave Ubuntu a thorough tryout for a month but finally gave up once I found out my WiFi printer couldn’t connect (as well as my graphic tablet and scanner - and MS FSX doesn’t run …). So … back to Vista.

The hair has received the most attention during the past few weeks when I’ve tried particles from weight painting on the head and separate rectangular meshes as particle emitters buried below the scalp. The RAM limitation during rendering is hopefully solved with ReallyBigRender, Vista BCDEdit and Save Buffers enabled under the Output tab. ReallyBigRender fails to complete the composite operation and the images don’t match up in tone when stitched together with GIMP.

Its taken a long time to sink in but I now realise that mixing SSS and AO is quite difficult and I am currently failing to use SSS on the eyeball and iris at present. I’m on the 148th model variant and 947 filesaves.

Attachments


Last adjustments on MakeHuman morphs and modified pose exported to Blender and matched with previous builds of eyes, teeth, eyelashes etc. More work on textures and hair particles. This last stage has 1.3M vertices and took over 10 hours to render.

Even though I’ve learnt a lot I realise I don’t know that much; my Blender apprenticeship is not over yet.

Thanks for viewing and for your criticisms.

Attachments