2.Game modeling workflows

Nowadays almost all games are using normal maps for ingame characters ,even mmos , so a high poly version of the character will most likely be needed to bake the normals for the low version. The normal maps can be done without the help of a high rez version of the model , from bump maps , even painted directly by hand ( i will show you some tricks i use to paint normal maps directly in photoshop later on ) though its less likely that you can achieve the same results as a high poly version will do.

The assets required for the creation of a character will most likely be:

• low poly version ( or multiple versions for different LODs)
• high rez poly version used to compute normals for the low poly one ( you could also bake occlusion maps and other maps that will help you later when texturing ...see paragraph "4.3.Transfer maps" for this )
• all kind of textures that the game engine will support ( diffuse, normal , specular are the most used ones)

I suppose that a concept/drawing is available and the steps that will follow are modeling/texturing.

There are mainly two approaches you can take when modeling ( the difference between them being the order in which the assets are created )

• low poly model -> diffuse texture -> high poly model-> normal map and other textures
• high poly model-> low poly model -> normal map -> diffuse and other textures

Both ways can give good results if done properly and the choice will mainly depend on the pipe-line already established; i used to work both ways in the past but i prefer the last approach.

Though the low/diffuse/high workflow might have some advantages ( like the low poly version of the characters will be ready to go down the pipeline to animators earlier in the process ,or the creation of the highrez version will be quicker if you work carefully and keep lots of selections into the diffuse psd file than you can later on apply as displacemets on the high rez mesh) it has one big disadvantage and i will picture it in a quick example.

Let us say that below is a piece from a model (a shirt with some buttons)... i assume the low poly and the diffuse textures are ready ...

The next step down the line will be detailing the low poly version so i will import a cleaned up version of the low poly into mudbox to upres and detail and i will also load the diffuse texture to check that the detail i will add will register correctly with the diffuse ... then start to add divisions/levels to the geo and adding details like big folds and big volumes ...

I will use height maps made from the diffuse texture to displace the fine details ( it will speed up the process and also this way i will make sure that the computed normal map will match closely the diffuse than if i will sculpt directly using the mapped diffuse on top of high ) ...

All is working well and i will export the final level to compute the normal map with maya/3dsmax or other program that will use raytracing for the process and will assure proper display of the normal map in the final engine but if i was not too careful when detailing i will encounter something like in the images below ( the effect was exaggerated here so i can make a point )

You can see that the details in the diffuse on one hand and the normal map on the other will not register correctly ... this is due to the fact that during sculpting of the big folds i displaced/shifted the geometry a little and though i used displacement maps to add details like buttons this will not assure that when raytracing the details for the normal map they will overlay perfectly with the diffuse ; to avoid this , the high rez detailing should be done with care if possible not big displacements from the original cage/low poly version ... but this will hinder you since the whole point will be to add as much variation in the surface direction as you can and big volumes also.

The issue will be toned down if you will use zbrush and will compute the normal map not with raytracing but with zbrush internal algorithm that will take into account all the subdivision levels , but its less likely that the normal map will be displayed properly in other software / engine if you will compute it that way , the old problem of seam display : every software i used to compute normals (and i have tried a few) have a slightly different method of computing/displaying normal maps in tangent space.

The best way for this is to compute the normal map with the software that will assure no problems in the end , for this project i used maya for rendering and i had to compute the normal maps within maya to make sure they will render properly.If you are using normal maps computed within maya ( with transfer maps tool ) and want to render with mental ray renderer within maya than make sure to check that in the Render Settings dialog -> mental ray tab->Translation->Performance you have checked Maya Derivatives (like in the image below - note that this option is off by default) this way you will use Maya’s derivatives calculation for bump mapping providing compatibility with Maya for mental ray, also note that this option is only available when Export Polygon Derivatives is also on.

ADVERTISE

The second approach for modeling : high poly/low poly/uvs /textures will give you the freedom to start sculpting and adding details from the beginning and you will get rid of the problem stated above also.
However ,when modeling the high version first you have to keep in mind that this will be used to bake normals for a low poly cage and model it accordingly.

You will have to estimate what details worth adding and that could be used properly when computing the normal map for the low poly version.

For instance if i will have the high poly below

...i could use a single quad plane for the low poly but the result will not represent the high version very good when viewed from above they seem too look identical

but when viewed at slanted angle you can see that the single guad plane does poor in reproducing the high detail

a better approach is to spend a couple more polygons to make the low poly better match the high version

then the low poly with normal map will look better viewed at slanted angles ( of course ,depending on the polygon budged you could add more geometry to get rid of the rugged look )

Also,you have to avoid modeling complicated shape details that will be hard to enclose within a low poly cage later on because they will "eat" too many polygons to be displayed properly ,though you might be tempted to go crazy and add lots of detail like shown below.

The high res model might look nice but when building the low version on top of high you will always have to stay within a polygon budget (5-10k triangles or whatever) and if you did not plan high poly details in advance you might face the situation where you will need more polygons than alloted to better match the already high model , and you will have to go back and change the high version.But with some practice you will know what detail worth incorporating in the high version and what details to miss because will be hard to reproduce.

Another big advantage of the last workflow is that you could also bake an ambient occlusion map and other maps (a simple diffuse pass or even specular passes to give more variation to the textures) using the detail from the high poly model and you could use them as a starting point for the textures.

Also you could try a combination of the two methods shown above, in fact this is what i have done for Varga and i will show you later on in the following paragraphs.