admin - February 11, 2010

My pipeline with characters is this:

1) I need a full color Front and a Back view drawing. Doesn’t have to be a T-Pose. I can model and make things look right from just looking at referance. If the character concept isn’t drawn and colored like this, I suggest that I or someone else do that. However, the drawing doesn’t have to be super duper photo-real even if that will be the final result.

2) I first model out the game model in quads till I complete it at a stage where the basic form and volume looks really good and has about twice as many triangles as the final game mesh.

3) I weld like crazy cutting the triangle count in half, making sure to minimize the amount of triangles. Doing things this way helps me to visualize the final form with minimal verts when optimizing. I try to keep the quad polygons as square as possible, this will give the normal map the best detail after sculpting the model in Zbrush.

4) I create the UV map using a checker texture, making sure the squares in the texture are evenly distributed over the model. This ensures that the model is getting the appropriate amount of pixels distributed over the model. I use traditional UV mapping techniques but with modern plugins to save time. I don’t use GUV tiles in ZBrush, that is not smart for games.

Here’s an example of my texture mapping style with lots of mechanical pieces:

http://www.jpwestmas.net/portfolio/kumo-3d

5) With a model that has a lot of hard edges I need A LOT of polygons to sculpt Nmaps with in Zbrush. A model that has a small amount of hard creases I don’t need as many polygons.

In the Kumo model I needed a lot of polygons to sculpt the mechanical pieces so I exported the model in separate OBJ pieces. I imported the pieces into a single Zbrush file using SubTools. That way I could subdivide each piece separately and get an overall polycount of 30 Million polygons or more in a single Zbrush file. You get a really nice Normal map that way.

6) I sculpt away in Zbrush using many, many different brushes and modes. I then export the normal map at a minimum of 2048 using Smooth UV, I can shrink it later if needed. I bring the Nmap into Photoshop and clean that up using painting and copy tools.

7) I save out the Nmap and create a copy of that. I turn the Nmap into a greyscale and colorize it. I use the converted Nmap as a template for the color and specular map.

In my 3D rendering application I apply the normal map to my game model and bake Ambient Occlusion into the recesses of the Normals. I bring that baked AO map into photoshop and apply that as my basic diffuse layer in the color texture. I then fill in flat colors in all the necessary areas using the Normal map as my guide.

9) I import my game model into a 3D painting package that understands Photoshop layers. That way I can tweak the look of the texture layers at any time. I import the layered texture and normal map onto my model so I can see it, tumble the model and paint on it in real time. At this point in my 3D painting I’m going for color/diffuse shading and not for details yet.

10) For fine detailing I bring the painted “color shade” texture back into photoshop for fine detailing. I can often tweak my converted monochrome Nmap to provide a lot of detail using multiply mode. I then paint opaquely over that with discretion.

11) I do a final render in my 3D application and paint my specular map by hand using the color map as my guide and previewing the result in my 3D rendering app.

12) Finally I test the model in the game engine and see how it looks with all it’s surfacing glory. =)