It has been some time since the last Ara related post on my blog. So I think it appropriate to give some overview about the current state of production and what happened during the summer.
Still back in June I had planned to have Ara’s Tale finished for submission for the Suzanne Awards 2010. During July I realized, that this goal wont be reachable without compromising motivation, health and professional as well as private life. So I slowed down and with much regret let go of the october date. Looking back it was a good decision, as it turned out that the summer was extremely demanding in regard to my professional life leaving not much free time and energy for anything else.
But work on Ara’s Tale has never really stopped. Texturing for the whole entry set is now finished. See here for a shot from the entry of the canyon back to the bridge
What came next were several appoaches to texturing Ara’s dress. This coincided with me trying out Mari. In the end, Mari ( besides too expensive for me) didn’t work out for me, because pressure sensitivity was not working with my setup. (Its a Qt 4.5 vs. actual xorg wacom driver issue, and Mari ships with its own version of Qt, so no fix possible here).
The work with Mari and several other texturing apps (Zbrush, 3DCoat) showed me what I would really like to do and let me build up quite some frustration with the current state of texturing capabilities of blender. Unfortunately the corresponding GSoC did not bring much leverage here.
All in all the last 3 months were a constant struggle with my inexperience, shortcomings in blender (2.49 and 2.5x) and not being able to get my visions implemented. I am in dire need of some feeling of success 🙂
Now that I can dedicate more time again to Ara’s Tale, I revisited my dress texturing approaches. Let me detail a bit how I try to do it and why and also show you the obstacles I had ( and still have ) to overcome.
To get details correctly done projection painting is the way to go for me. Until now I almost never had used blenders projection painting for texturing but had always directly painted in gimp, but I was never really satisfied. If you could solely rely on projection painting, uv unwrapping could be much easier, as you would never had to go to the 2d image for painting.
For the cliffs I mainly used the clone tool with cloning from another uv layer to get the rock textures placed. With texturing the dress I needed a much finer control, meaning to really paint on the dress. This very soon showed the shortcomings in blenders paint engine. A very limited brush engine and way too few blending modes made me soon recognize that an external painting program was still very needed.
To get the needed level of detail for the dress I split the UVs into patches for the top, the skirt and the belt.
This is also a leftover from my tests with a combined workflow with Mari. I actually have 3 uv sets, the 2 last sets just copies of the first shifted to the left, so that the affected vertices are at the center.I will than have 3 materials for the dress which will use these uv sets.
Now for the nice part blender is offering. For a slightly better texture painting support I use blender 2.5x for the texture painting. What may be not so commonly known is, that you can assign different images to each of the faces. For the diffuse color map for the whole dress I assign 3 images to each of the 3 parts of the dress. When entering texture paint you can paint on the whole object and will effectively draw on 3 images at once. These images may have different resolutions. This is a very convenient way to draw textures in a consistent way.
For the actual painting I use a combination of the ‘clone from another uv layer‘ method, direct painting with simple brushes and sometimes using the quick edit feature, where the current projection is exported into an external editor ( gimp in my case). You can add details there and apply these changes back to the texture image. Its important to note, always to use the exported image as a reference and do all the extra editing in a separate layer and only apply this layer back to blender, otherwise you may get weird artefacts or double coloring.
After some time I got used to the workflow and with patience I actually got the work done. This has to repeated of course for all the different channels (bump, spec etc). A better and less tedious setup process would be a real help here.
The first test renders revealed a problem with the bump mapping implementation in blender 2.49.
Consider the following setup
A simple uv sphere with a seam at its front and simply uv unwrapped. In texture paint mode some gray strokes are put so that they cross the seam. The painted image is used as bump map for the sphere. The resulting render looks as follows:
Note the very visible seam. That discovery is just one example of the many obstacles I met during the course of this project.
And here is the workaround:
Blender 2.5x does not exhibit this problem, but I cannot use 2.5.x for rendering as its does not import 2.49 files properly or not at all. The solution is to setup a material in 2.5x with the created bump map. I can then bake a normal map which is essentially a conversion from the greyscale bump map to a tangent space normal map. Using this normal map back in 2.49 works perfectly – no more seams are visible.
Its working, but introduces a huge overhead and potential error sources, but I know of no other solutions here.
The good side is, that I can finish now the dress texturing and it only involves work but not magic. Its also some valuable lessons learned which will be handy for the next texturing job – Ara herself.
With all this recent experience I find it hard to predict a definitive release date for Ara’s Tale but hopefully it will be early 2011.
And I am pretty sure there will be a Suzanne Award 2011 as well …