I have the visuals done for scene 1, 15 shots in total. There may be minor tweaks along the road but nothing serious any more.
Finishing the shots for scene 1 was a testbed for me to discover the needed workflow and techniques to bring the shots to life. It wasn’t clear that I would ever be able to achieve what I had envisioned, but to my delight I am very pleased with the results.
This shows me that the short can be done (with a lot of work of course) and that gives me the motivation to keep on working. Further on I have now the techniques in place to tackle the remaining ~45 shots in an efficient way.
See here for some frames taken directly from the shots.
And here are the first 103 secs of Ara’s Tale. 3 shots in the inner-eye sequence are still from the animatic as they require set 2 to be ready, which it isn’t yet. The shots were rendered at 50% resolution (960×540) so you can easily scale the video up to see more details.
I will now slow down a bit after the heavy workload (and long nights) during the last few days. After that its time to start the modeling and texturing of set 2 and sculpting and texturing the dragon.
That means that updates will most probably be a bit less frequent for the coming weeks.
I also won’t show full shots from now on, but only stills, but I hope to get a trailer/teaser ready sometime early next year.
I thought this would never happen, but slowly something like a production routine is taking hold of Ara’s Tale.
For the past days I was busy working on finalizing 5 more shots and could actually reuse knowledge gained from the previous shots. I now know for the current setup how to achieve the desired effects and how to setup the lighting and needed render layers. The same is true for working with the compositor, where I can now reuse a lot of node groups and techniques created so far.
That doesn’t mean, that its now an automated procedure, but there are no detours and breaks due to unwanted surprises along the road ( … so far ). I can now concentrate more on the look and feel and not bother so much with the technical aspects.
I plan to do 5 more shots and then release a preview of the first 80 secs, which I think will conclude the publicly shown shots.
Depending on the material, there might be some positive effect of doing a teaser/trailer. This could be used as test bed for the collaboration with my sound and music experts, but that really depends on their time schedule and possibilities.
Here are frames from each of the 5 shots I have done the last days. These stills are taken directly from the shots. The resolution of these shots is currently 720p.
As announced in my previous post I present here a shot compositing breakdown from a frame of shot 3 of Ara’s Tale.
This is a medium shot with emphasis on Ara and the environment does only act as nice blurry backdrop to the frame. So the main focus lies on the foreground with the background representing one renderlayer with a simple light setup (1 spotlight with shadow buffers for the moon and 2 hemilights for the ambient lighting).
In this post I try to portray the things I have learned and experimented with about combining image layers having both transparency information (alpha) as well as z information.
The starting point for me was quite simple. Up to now I had done the compositing work for the background, Ara and the cage. As I had not yet settled on the exact look of my cage lights I decided to add them later. This shouldn’t be so hard after all.
Sometimes it seems, that the sole purpose of this project is to test my patience, motivation and capacity for suffering.
Setting up a workflow for offline compositing, ( i.e. compositing not directly after rendering but from saved multilayer exr files ), sent me on a journey through storage demands, SSS deficencies, the current state of blender 2.5, back to 2.49, down into the source code of compositor nodes and finally to a somehow (shakily) working solution.
After fiddling with the skin shader and lashes for Ara, I thought it a good time to start to experiment with some lighting and compositing work. The main idea being to have an environment to test the different shaders for Ara and to spot some possible problems.
I chose the very first shot as test bed for my experiments.
I use all linked assets into my scene files and the last addition to Ara were her lashes – as separate objects – so I linked those as well and … first surprise ! The lashes imported but never linked correctly to my armature proxy. After some fiddling I decided to completely reimport Ara and associated objects. I have the feeling that I might might be doing this more often in th future, so I made a quick howto for me to not forget all the small little details to get it right.
Next came the baking of the cloth simulation. I had already invested quite some time into this topic, so it should just be a setup-click-and-wait kind of thing.
Ha, so wrong !
I the past I had concentrated on the skirt part of Ara’s dress and quite neglected the upper part. Now I had a wider view and recognized that the current setup just isn’t working. This insight resulted in a separate collider object for the dress, a reworked weight-painting and remodelling of the dress mash and dress colllider as well as a slightly different setup for the cloth simulation itself.
Its better now, but I see now that I will have to refine this from shot to shot.
Ok, next came the lighting setup – no real surprises here (for the moment) – and accompanying the compositing part.
The topics I wanted to cover were
separate layers for each light group (main, ambient, moon, cage lights ..) for each part (background, Ara, cage) to have maximum freedom for balancing the lighting
vector blur for all objects
DoF
Play around with some nice glow effects
Find a way to light a night scene with a path in the shadows of moonlight were the main light source is coming from a cage.
Well, the first point in the list was easy. The funny things started when I tried to use the vector blur. I have Ara and the cage on different render layers and combine them using the Z-Combine node. Well when using the vector blur ( or any other blur node) this will not work correctly anymore, as the z information is sharp and not blurred at all ( and this isn’t possible anyway). I have not yet a solution, but I have to somehow create a mask originally derived from the z-buffer of both layers and apply the same filter operations on this mask as well. At least thats my theory for now.
Pure alpha-over will not work, as I have intersections ( Ara holding the cage).
Another thing which is bothering me is that I do not seem to get a vector blur on my background, which is realized as set ( set in the render panel, where you do not have any selection possibilities). I have to investigate this further.
Another immensely annoying thing suddenly popped up again: all seams on Ara’s dress were clearly visible again. I thought that baking the normal map from bump maps in blender 2.5x was the solution for this ( see this post for more info ). As it turned out, this is not he solution, as blender 2.5x has the same behaviour.
After long ( very long) experimentation I finally found the culprit which causes this:
It seems (no pun intended) that around seams blender is using areas of the texture map beyond the actual uv-mapping, maybe for blending/smoothing purposes. The default setting in texture paint mode for the projection painting is to have a bleed value of 2, which means that blender will paint also 2 extra pixels outside the uv-mapping. This value is ok for color maps but obviously too small when drawing bump maps. An increase to 4 or higher solves the problem – in principle.
What it means for me is that I have to repaint all seams in the bump maps on my relevant models with a higher bleed value.
The DoF point was another source of new work. In the early animation phase I created a camera rig with all nice controls for positioning, targeting and controlling the lens value. What I didn’t think of at that time was to include a control for the focus point as well … Ok, time for a new reimport session for the camera rig.
And finally I quite clearly see that a lot of polishing will have to go to the animation done so far. But that was to be expected.
All in all a very informative and in the end productive weekend. To show you the outcome of all these explorations I have uploaded a 4 sec shot of the shot. It’s a very early glimpse of what is to be expected and still no vfx ( glowing lights) done here. The irony here is that in this clip Ara’s hands are rendered black due to rendering time constraints.
After finishing the skin shader one of the last things left to do was to create Ara’s eye lashes.
From my first research on hair it was relatively clear that the eye brows and eye lashes would have to be implemented using textures. I did the brows already during the development of Ara’s skin shader.
For the lashes I started with polygon meshes attached to each eye lid. Alpha mapped images of the individual lashes were applied via appropriate uv mapping.
It did work, but not really to my liking. Using this technique I created lashes which were essentially 2-dimensional ( aka flat). And not only this, but the roots of each lash hair were in a more or less straight line. This is ok if Ara’s is seen from a distance but gets increasingly annoying in closeups. And said closeups definitely make their appearance throughout Ara’s Tale.
Karl Nyman (chipmasque) mentioned that he created lashes using 3 different techniques and his favorite was implementing the lashes using a particle hair system. After some discussion about the scaling issue with hair in blender I nonetheless tried the hair approach.
This is the parameter panel for the strand settings for the hair shader. As the input fields for ‘Start’ and ‘End’ suggest, it should be possible to enter values to 1/1000th of a blender unit. As my previous tests showed 1/1000th blender unit was just 2 to 3 times bigger than lashes for Ara’s would have needed.
In a total rush of optimism I tried to input my needed value 0.0004.
And …
SURPRISE
the values are correctly interpreted, ts just the display values which get rounded to the nearest 1/1000th.
This made things different. I setup Ara’s main hair with blender units using the ‘hidden’ scaling feature and what can I say: I just worked. Not only that I worked better than using the pixel based approach, as now the hair is always displayed in the correct width, regardless from how far away the hair is seen. I still use the child simplification technique to lower the rendertime with no ill effect so far.
The only drawback with this method is that you have to memorize somehow the actual values you entered. Because the next time you go he the input field it is reset to the nearest rounded value. Further experiments showed that the input routine accepts and correctly interprets values down to 1/100000th of a blender unit.
This is absolutely cool !
Another topic since my last post were shape keys and their intrinsic behaviour. During a discussion in my wip at blenderartists I realized that a lot of the corrections I made to Ara’s base mesh weren’t visible in the renders. It turned out that my understanding regarding shape keys (morph targets) was wrong.
A big thanks go to kram1032, who suggested to introduce a new shape key which holds all the corrections to the base mesh and is always fully visible. This trick is simple and just works wonderfully.
The image on top shows the current state of Ara with all these new techniques applied. And because its just nice to see here is a link to a Full HD version of that image.
Whats left to do for Ara is texturing the hands and feet, which I already have began. The shader will be simpler than the face skin and I do not expect any surprises here ( now why does that sound odd … )
ps: after seeing sintel ( very impressive: storytelling, cinematography, cg technique; a very satisfying experience) I am now even more motivated to finish Ara’s Tale.
In the comments section of Ben’s post are some interesting links to papers giving quite a good insight into this problematic area.
The basic setup as described in Ben’s post is here too, but after a lot of experimentation I expanded the nodes a bit:
I use an additional sss mask for blending a sss version of the subdermal sss layer with a non-sss one. This gave me better control for applying the sss effect.
I modified the specular term by simulating a type of fresnel specularity. In my opinion this made a huge difference in the skins appearance.
It also took quite some time to develop a workflow for texture painting. I did a basic material setup in blender 2.5x with the 3 layers as texture slots and using the same mode as in the original material setup. Using GLSL materials I could than texture paint relatively easy.
But due to the fact that the 3 layers are combined using the screen mixing mode it took some practise to find the right colors to use. To have at least a bit of a non-destructive workflow, I used various layers for the skin details.
Interestingly the biggest problem in the end was to apply the brow texture. Usually you have a texture map with just the brows painted on an alpha only map. As texture layer you just mix it with 100% on top of the other textures. With the multi material setup where all the layers are cumulative I did not find a way to do this easily.
The node setup right now demonstrates for me that this approach is working. On my todo list for the skin shader is:
convert all color operations to a linear workflow correct version (screen -> add, …)
clean up the huge amount of work done by all the extra texture layers by combining them into one or baking normal maps.
The image shown above was rendered at 75% of the final resolution ( 1920×1080) and was slightly over my limit of 3 min per frame ( 3:15). I hope to reduce the render time with some of the cleanups mentioned above, but the biggest part took the hair anyway (~2 min).
I have to confess, that I may have overdone it here in regard to realistic display. I have to render a shot to see if the animation doesn’t look too clumsy now, in which case I even might consider tuning down the skin shader.
And btw, if anyone is wondering, the lashes are still missing.
I started to texture Ara’s face. The shader itself is only a plain lambert shader where I use some color maps to preview the painting result.
To have some guidance later in the texturing process, I rendered out a test render with the current state of texturing and than I started doing paintovers in gimp. These experiments are not finished, but I wanted to show one result of these tests.
I will use some sort of multi layered SSS skin shader but I have to experiment a bit with these to see how to approach the painting of the different skin layers to get the desired end result.
). I can now concentrate more on the look and feel and not bother so much with the technical aspects.
