Here is a small testshot of the entry scene highlighting the areas of work I did during the last weeks, mainly cloth simulation, hair and environment texturing.

The lighting used here is a bit more than basic and gives a better idea of the mood for this scene. But its only simple lighting, no render layers and no compositing or color correction.

Cloth simulation looks ok now and will be used for the final. The hair looks ok but still needs some more attention, but I am relieved to have it somewhat working. The environment texturing is close to final for the principal textures, only some more subtle coloring will be added.

Whats still missing is texturing/shading of Ara and her dress.

Right now one frame from this shot has a rendertime of 56 seconds in full resolution (1920×1080) on my amd quadcore system. This makes me optimistic to reach my goal of ~4 min per frame. An Interesting side note here is that the preparation time before the rendering itself is starting is 30 secs, so the actual rendertime is 26 secs. As the preparation time does not take advantage of multiple cpus this is a value which does not scale with more cpus.

Right now my time tracker is at 736 hrs and still counting:

5:34 01-Script
43:54 02-Storyboard
73:54 03-Concept Art
14:48 04-Animatics
121:20 05-Modeling
76:04 06-Rigging
171:30 07-Animation
65:21 08-Texturing
54:51 09-Simulation
10:10 10-Project Management
98:53 11-Research
----------------------------------------------
736:19 Total


Its good to see all the single parts put together into a small shot. Whats even better, is that it gets exactly the mood and look I had in mind when I started this whole project. This is very satisfying and helps bear the sometimes difficult times during production of this short.

The shot was rendered at 75% of the full resolution, so do yourself a favour and view it fullscreen.

I decided to write a post about the current state of texturing the rocks and cliffs for scene 1.

Although far from finished it is a first approximation of what I am aiming for.

After the various expeditions into cloth and hair land I thought that texturing was a voyage into known territory. As like so many preconceptions during the development of this project, this turned out to be some wishful thinking again.

The topics I encountered so far a many and quite diverse. Currently I am at iteration level 4, with twice a complete modeling restart and 3 scultping reworks.

Its a delicate balance you have to do between the amount of modeling, scultping and texturing. Doing it wrong just gives a fake feeling for the cliffs when viewed from different distances. And due to rendering time limitations I refrain from using extremely detailed displacement maps ( with accordingly high poly count cliffs) and try to cover this with bump/normal maps.

I also do not longer sculpt the high details for normal map baking. This could work out if the texturing were painted exactly to the fine details, which turned out to be to much work compared to what is actually gained.

For the scultping I use the GSOC branch from J. Wilkins available from graphicall.org.

The texturing is done with the current blender 2.5 svn trunk. It took several attempts to arrive at a suitable workflow. With my previous projects, I never seriously used the projection painting system from blender. For Ara’s Tale though I wanted to use this feature.

I ended up with a mixture of projection painting and working in gimp, which unfortunately is far from efficient and even leads to compromises I didn’t intend to make.

When working with graphics I massively make use of layers and masks and try to implement a non destructive editing workflow. A typical use of layers is to test things, play around with blending modes and keep different versions in one file, being able to go back and switch between texture approaches.

Alas this is a non-option with the current implementation in blender. Whats even worse are the limited painting modes, with the missing overlay mode the most prominent. The brushes are also a bit limited.

A way out of this would be the ‘Quick Edit’ feature but I had no luck to get it working in a consistent manner. Reapplying the external edited projected image always resulted in either a tinted texture (which I was able to trace back to hemi lights active ?!?) or just a darkened image. So multiple uses of this feature leaves you with stripes and too dark an image.

What I do now is to use the projection painting to layout the basic texture and work on the seams. It works quite well to almost erase the border seams between different objects. The resulting texture is loaded into gimp and here he high frequency details are added. Back in blender the seams get cleaned up again and some stretching from the gimp editing is corrected.

For the cliffs I decided to use grey scale photo patches. With this I am able to design the basic structure. based on the greyscale texture I start coloring and have a base for the bump maps.

I am using photo textures all taken by me during my last visit to arco/italy.

See here for some photos to get an idea of the environment there. ( lighting quality is bad from a photographic point of view, as I intentionally went to places which were in shadow to get only diffuse lighting, which is best to take textures )

I took some 150 high res RAW images which now is my base for all the various rocks in Ara’s Tale. All were converted to greyscale, a bit sharpened and lightness corrected to have a consistent texture base.

An important issue was to keep the same scale across all the different cliffs. I used a reference object as starting point for the zoom factor and subsequent project painting. It also helped to have a model of Ara there during painting to fit the rock texture the scale of a human being.

All this said I am still not 100% satisfied but anyway here are some samples of various cliffs of scene 1. The lighting varies over a wide range and hints at the next major headache. Right now its a bit on the dark side and you may have to have a properly calibrated monitor to see it correctly.

And finally here is a shot with some textured cliffs to show how this all works when animating.

Oh, and I almost forgot, I also did some work on the texturing of the cage.

Setting up cloth simulation is a piece of cake.

Really.

You should try to setup a workable, reliable and usable hair simulation and you will agree with me.

[warning, rant ahead]

Static hair setup in itself is a tricky endeavor, but can be handled with patience, some know how and the will to make compromises. One of the many compromises e.g. was to stay away from curly hair, although this was high on my wish list. Well I could just comb the hair to get curls, but I really want to have the movie finished in my lifetime.

My last post showed the preliminary results of my static hair experiments, with the almost prophetic statement, that this may not stay this way once animation starts. With animation I mean hair with softbody simulation to give a nice fluid movement and good interaction with wind.

Quite frankly, hair with softbodies ( in blender 2.49) is a nice demo feature for carefully tweaked shots, but just forget it for serious work, especially with long hair. This system is so sensitive to what you do and in what order.

Every so often I got just totally messed up hair after a few runs of softbody simulation. Sometimes it was just enough to move the actual frame on the timeline. Tweak some hair parameters while the softbody simulation is still on and you get a 50:50 chance for a total mess up. I had 4 times where my basic hair setup was totally messed and I had to start from scratch again.

This all is accompanied by the fact that hair softbody simulation with collision detection (not self collision) is almost a magnitude slower than cloth simulation and stretches your patience even thinner.

After a long time tweaking I somehow got a very shaky and tedious workflow going and went on to the next step: adding wind.

HA!

What a heap of ….. !

Wind alone is immediately visible on the hair without the need of any softbody simulation. So its nice to adjust the strength. But what happened in my case was that the force direction of the wind changed by 180° and increased in strength once the softbody simulation was activated, but only for the hair of course.

I still have no real idea what exactly is causing all these problems, but it got me immensely frustrated. So much in fact, that I consider skipping the long hair for Ara. If 2.5 is not better with this I really do not envy the durian team with their task. Bick Bug Bunny had it better as they didn’t use the softbody simulation.

And in an incredible masochistic move I tried to use the hair system for the brows and lashes. For full control over the shape of the brows and the individual lashes you must not use children here. But with children disabled you have no longer the child simplification which is necessary to counter to hair scaling problem. This results in extremely bushy brows the farther away you are from Ara.

Conclusion: brows and lashes will be painted.

[/rant]

If anyone has some good advise I am all ears.

In the meanwhile I will try to keep the long hair and trick, hack, bite, weep, shout and smash the hair for each shot in place.

( … and I may book a nice room with soft walls in our local psychiatry … )

Just a quick glimpse at what I am aiming for regarding Ara’s hair.

I have not yet tested if this setup and approach is applicable for softbodies and/or animation but I thought I share it nonetheless.

While trying to get everything in place for the entry scene I also allocated time for some serious cloth simulation.

Some may remember that I had several attempts to setup a workflow which would allow me to do fast animation prototyping and yet have all the links to the cloth simulation ready.

Well, that was my impression some days ago too. And then I started to actually apply the cloth simulation to one of the entry scenes. It failed miserably.

I had overlooked one little but significant detail. As suggested in several tutorials and also the blender wiki, I went for a setup with a combination of armature driven (torso and upper arms) and collision driven (skirt, sleeves) type of cloth simulation.

With such a setup it is important how you define your starting position. Usually I start in the rest position and slowly ( ~ 30 frames ) settle to the first pose before starting the actual animation to give the cloth simulation time to settle in. As parts of the cloth are only collision driven you have to find a way to fast position your cloth. My setup so far was to parent the dress to the main position bone of Ara’s armature. In each scene I only modified the position bone itself to move Ara ( and the dress around).

As soon as I actually enabled the cloth simulation and armature modifier on the real dress ( animation was done with the fake one) things got messy. Moving the position bone moved all the parts of the dress correctly which were only collision driven, but all the armature driven parts moved twice the way (of course) .

The correct setup would have been to position the armature object at the animation starting position and leave the position bone, which is not the usual way of doing it.

So now I had to manually reposition each animation to be based on the armature’s object position instead of the main position bone. A lot of tedious work. And this was now only for the first scene.

Anyway, I found a way and started work on the cloth simulation parameters. And that cost me ~30 hrs of work to arrive at a solution which is satisfying.

There are some (few) tutorials and threads on cloth simulation, and I also have Tony Mullens book Bounce, Tumble and Splash, but there is nowhere a real in depth discussion on how to best tackle cloth simulation. And as I see it, there isn’t only one type of cloth simulation but a whole spectrum full. With the sparse information you are more or less on your own and have to figure out how each parameter influences the simulation in this specific setup.

To have the turnaround cycles fast I lowered the simulation quality. This turned out to be a major mistake in my case. It seems if you have a very free moving cloth system where energy is easily added ( skirt collides with legs and gets accelerated or stopped ) the simulation tends to oscillate and isn’t able to converge on a good solution. What I observed was that as soon as I started to get good approximation of a dress fabric, the skirt started to get crazy after a few steps.

In a last attempt I increased the quality setting (from 5 to 20) and was amazed to witness a totally different ( but good) simulation). With this I could now really start to map out the best parameter settings. With 5 secs per frame and at least 100 frames before you have a good estimate on the type of simulation it was a long and tedious work.

Its amazing how many aspects contribute to a convincing cloth simulation. First you have to get the parameters right of course. Then the part which is armature driven has to be fine tuned, otherwise you get some strange artefacts. Stiffness and bending scaling was an important tool. I wished there was a mass scaling as well. And last but not least the shape and topology of the dress itself was equally important for nice and fluid movements. I got some hints from actual dress cuts of a fairy dress for my daughter.

What I think is not really achievable is a very stiff cloth skirt, which doesn’t bend easily and keeps its form or bends in sharp edges.

I have now a setup which is so far tested in each shot for the entry scene. The trickiest part is where Ara’s is walking and I am confident to have all aspects of the main scene covered as well. But that will be seen when I start work on that …

Oh, and yes, I still have to test it with some wind effectors as well …

After so much babbling see here shot 3 with the actual cloth simulation in action.

I wonder what surprises the hair simulation will have for me.

A small update here.

I worked a bit more on the eyes and started to put some facial features on Ara’s face. I used the current SOC sculpt branch from graphicall for the sculpting and I am quite pleased with the tools which get developed there – very great improvement going on here.

After the still image from the last post I wanted to see the eyes in action and since I have a shot where you see an extreme closeup of the eye I put some crude lights into the scene and got the shot rendered.

There is only the normal map active an Ara’s face and a basic lambert shader for the skin. The eyes are fully textured now. The lighting right now is a pain and you better ignore it, but I didn’t want to go into the lighting just now.

I am going to rest the project for the coming weekend as I am going to visit Arco in Italy, which is a great place for climbing. I will not do any climbing myself, but the rock walls there are spectacular and coming just right for building up my texture library for steep canyon walls.

I will post some images when I am back again.

After all the sculpting and modeling for the first set I thought is was time to enhance Ara as well and work on some remaining tasks such as the eyes.

Until now the eyes were pretty much a placeholder for the real ones. The outer geometry was there to aid in the rigging and shape key definition, but no further modeling/texturing was done.

As the eyes are said to be a mirror of the soul I wanted Ara to have eyes that could accomplish this. That meant no pixar style eyes and after browsing through a lot of eye tutorials I finally settled on the workflow of an eye tutorial from the gnomon workshop. Its for maya, but he principles are easily transferred to blender.

Well, the modeling part was quite fast, but the texturing part was a little bit more time consuming. And its also a bit too early to start the texturing, but in this case you have to texture it, other wise you cannot see the iris and thus the most important part of the eye.

During my render tests for the image you see at the beginning of this post, I soon realized that the lighting task will be very important to bring the eyes to life. I also ran into a problem with the render engine. I wanted to have raytraced refraction activated on the cornea of the eye, but if I do this, the bumps on the iris are not rendered. Right now I use simple ztransparency to get the bumps, but the eyes do not look good from the side. I also tried the standard render engine from 2.5 but got the same result. Looks like I have to find a compromise somehow

Another point I notice now that the eyes are more defined, is that I have to work on the alignment of the two eyes. Right now they seem a little off.

And for those who are wondering, yes, Ara will have red hair and green eyes

Having established a suitable workflow I went on to apply it to the rest of the set … just to run into the next wall.

I still had the view of building a prop like in a still. But in a movie the same prop is seen at various angles and distances. Sometimes different parts of the same object are visible in individual shots are very near distance, meaning they take up a lot of screen space and thus need to have a high level of detail.

I tried to attack this problem by splitting up the uv map with the high detail zones taking up a lot of map space. This cannot be applied to scultping of course as I do not ave the scultpris type adaptive subdivision.

That meant I also had to split some props into several parts each with its own maps and uv maps. Splitting an object into more parts introduces problems when you have to ensure a continous texture across the now individual parts.

In at least one situation I had to build a second stripped down set with just a very simple environment object, which will be seen at very near distance. The environment may have motion blur and dof applied but right now I cannot exactly estimate how much detail is still needed, so I try to keep on the safe side.

I have not used other packages than blender, but what I have seen so far indicates, that in other software you are able to apply individual texture maps to uv sub patches of one model. Using a good projection paint system you can texture models to a very high level of detail without loosing the control of the overall coherence. There doesn’t seem to be similar technique available in blender, so I have to resort to high resolution texture maps ( 4K+), model splitting and uv patch tricking.

Anyway, if anyone here has a good idea on how to handle high resolution details on different parts of the same model I am more then interested in hearing them.

In the end the entire set modeling/sculpting turned out to be a highly iterative process. I have now a growing list of small todos to get everything in a state I am satisfied with.

I think I will continue with the texturing and maybe even cloth and hair simulation all for the first set. This should give me a solid foundation to tackle the next set, which isn’t more complex but has a lot more shots and may pose some hard questions. I hope to have found most of the answers before advancing the the next set.

I have rendered out 4 shots with the current state of the modeling/scultping in place. No color texturing and no lighting which deserves this name are in place.

It may be common knowledge, but for me its my first movie project and I am just amazed how dramatically the shot changed by just substituting the blocked environment with a fully modeled and sculpted one. The changes are so big in fact, that I may have to revisit some of the camera settings in the other shots to better deal with the added detail. A very intriguing process.

I wonder if the effects when applying the textures, proper lighting and compositing are as huge as this one.

I have put together a video which shows shot 3 simultaneously in the two versions.
Keep in mind that only basic modeling and scultping was done. Lighting setup is nearly non existent, just to be able to see something

This turned out to be much more time consuming than expected.

Modeling and sculpting are very near in my workflow. I create a base model and sculpt the details, which will be applied back to the model using normal maps. From the scultping process I also derive an AO map for direct inclusion into the model’s shader.

As I have already mentioned in my last post, I evaluated some tools to ease the process. I tried another attempt at using scultpris, but it just doesn’t fit into the workflow and it still has a lot of issues, crashes and quirks which in the end will cost me more time.

Using the bridge from the entry scene I tried several approaches on how exactly to implement this modeling/sculpting and texturing workflow.

Right now I have arrived at the following:

• I create the base model in blender 2.49 with all the basic rough geometry in place
• an uv map is created in blender 2.49. I create uv patches where those areas which occupy the most screen space also occupy the most area in the uv map ( sort of a cheap handmade ptex .
• the model is then imported into blender 2.5 and a multiresolution modifier is applied up to the resolution I need for the specific purpose. In case of the bridge I used up to 2 million vertices for the scultping
• Then starts the scultping, which is way better in 2.5 then in 2.49 but still no real match to zbrush, 3dcoat or the adaptive mesh tesselation in sculptris.
• From the sculpted mesh an ao map and normal map is baked into separate images
• The generated maps are then used back in 2.49 for the texture slots in the shaders

See here for an example of the bridge and how the individual maps apply:

Basic Modeling

AO map applied to base model

AO and normal map applied

I would love to have the depth painting features of zbrush, 3dcoat or even sculptris. As you can see I use the scultping only to generate my normal maps, so to be able to directly paint in depth would be a real blessing.

I already did some further investigation into texturing. What I will do is the follwing:

• Again texture painting is done in 2.5 because the tools are superior to 2.49
• I create a color texture slot for the material with an empty grey image map loaded
• I use the GLSL feature to display the full material including ao and normal maps. This gives me a good guidance where to put the texture.
• Well and than he painting starts, with heavy use of the clone feature
• Once finished and if the targeted material allows it, I temporarily use the created color map to act as bump map too.
• With the previous normal map active and the color map acting as bump map I bake a new combined normal map, which now include the high frequency details from the color map as well.

In my next post I will show you the result for the bridge.

Here are another two examples of the bridge in two other shots.