Recently, Ben Dansie posted a thread on BA (at the time of writing, the BA server has been down for the past day or two... perhaps at some point it might come back) reporting his progress in making a "lite" version of the Sintel rig from the Durian Open Movie Project. This is currently at version 2 (EDIT: link here - 11mb Zip).
For anyone that has played with the original files, the "full" sintel.blend that was used actually used in the production was a monster of a mess. Apart from having some rather heavy-duty geometry accompanied by heaps of high-res textures, there were so many different objects and versions of Sintel in the file that it was really hard to tell which one to use (unless you were a "core Durianer" I guess ;)
All of these factors mean that on the computers of us mere mortals, making use of these wonderful resources would be a darn lot harder. But thanks to this great initiative from Ben, hopefully this will become more accessible to a lot more users while acting as a good "from production" test sample.
Today I had some time, so I decided to have a little play around with the file, testing how expressive the rig was, as well as practicing some animation skills.
Opening up the file and rendering it, the first thing I noticed was that the hairstyle looked quite "bleh", especially compared to the shots in the film. So, I decided to give the particle hair combing tools a good test drive (for the first time, not on a sphere or a cube!). In this pose above, I initially just started by fluffing up the hair a bit, before starting to experiment with adjusting the length.
At first, everything was going quite well, with the relatively short(er) hair. But as I started to make a longer, more flowing hairdoo, I started running into some increasingly frustrating problems:
- the comb brush at times started to mess up stuff I'd already gotten all nice and pretty, while not tweaking what I had intended to fix
- some stray hairs started getting a life of their own. No matter what I did, I could not manage to comb, cut, or otherwise wrangle them into a nice place
- perhaps my particle-hair combing fu is lacking a bit, as very soon into the process, I started getting "bald" spots on the top of the head, where the hairline abruptly cut off. Trying to fix this of course was frustrating, as at times I couldn't get the right bunches of hair to move to cover things up, without messing up what I'd already set up. Grr!
- toggling between particle edit mode and object mode, I noticed that there were some rather significant discrepancies between the hairs shown in either mode, with some of the "guide" hairs (I'm guessing about the terminology here) being in places without any real/visible hairs, or some weird hair clusters where there were no guide hairs in sight to try and use to fix the woes.
- trying to get a tight-fit on top of the head, with some contours proved quite difficult, especially after trying to avoid bald patches and/or after adjusting all the hair to try and fix the outliers...
Having finally gotten some hands-on "hair" experience, I can say that we're actually still quite a way away from having truly optimal/nice to use hair tools. And I haven't even tried to run sims on these yet! Gee... I can only imagine the pain of working on a character with several hundred feet of hair. Not to mention the pretty-common in real-life "hair-tie + hair + hands" type interactions, which AFAIK would still be a nightmare with the current state of the art tools. (Hmm... perhaps this area warrants some further investigation...)
Now, moving on to the rig. Having played around with some of Nathan's older rigs, IMO, some aspects of this rig feel a bit clunkier/less intuitive/comfortable to use, at least when getting familiar with the rig.
One of the first stumbling blocks for me was that Auto-IK on the arms doesn't work due to some of the trickery used to implement some other features (i.e. "hinges"). This made posing the arms a lot more work, since I'm used to just roughly pulling arms into place before carefully adjusting angles/etc. for a nice fit. I could've gone straight for the IK arms instead I guess, though with those you do lose a bit of fine control later on...
The other main issue was the facial controls. While I'm aware of the time constraints and compromises that occurred due to these during production, I'm not convinced that the reliance on custom properties on fixed bones is actually such a great interaction design. IMO direct manipulation of bones (or widgets) on the face feels more natural, and introduces less clicking/mental overhead required for remembering which bone(s) had which properties to do what, as the controls are situated closer to what they control. Having said that, I'm not too fond of those rigs which use "solid" widgets (i.e. shaded custom bone shapes), as these give the rigs a really tacky "plasticky" clunkiness, which doesn't really do animators any favours when they're trying to create any quality emotive-poses.
Speaking of facial controls, in the pose above, I was going for a "sneeze" expression, though perhaps the lack of a nose "scrunch" control didn't really allow that to get conveyed that clearly... As a result, this pose probably looks a bit more like snoring ;)
Alrighty, enough blabbering for now. It's time to get some sleep. Maybe tomorrow I'll have time to make a few more... Until then, you're welcome to post critiques/interpretations of the poses, or even suggestions of poses to try!
Is it wrong that I think this new version looks way better than the official sintel.org version.ReplyDelete
Really excellent experimentation.
great work on these pics. really great poses.ReplyDelete
Thank you also for all your work to better blender with your awesome commits!
I think the hair gives too much personality in CG, so changing it changes the whole character to the point she doesnt look like sintel.ReplyDelete
Btw, and sorry for going of topic, can someone tell me where is the latest code of bullet build (GSoC 2010)? the latest version for linux in graphicall is from 8th August. I could try to compile it to have a more recent build. Thanks
As a "core durianer" I can tell you that the official Sintel is no easier for me to navigate.ReplyDelete
Nice poses, BTW!
Nice! But what's with the green shadow?ReplyDelete
"I'm not convinced that the reliance on custom properties on fixed bones is actually such a great interaction design"ReplyDelete
Agreed, of course. They are simply an easy stop-gap until I can figure out a way that I think is actually good.
The drawback of on-face controls is that, although they are indeed intuitive, they are ill-suited for more complex or anatomically-based face rigs. For example, rigs based on FACS. The reason being that many facial expressions are the result of opposing muscle groups activating at the same time, and this is difficult to represent intuitively with grab-n-go on-face controls. I tried to make this work for Sintel, but eventually gave up.
I'm beginning to see the wisdom of off-to-the-side box controls that are so popular for many face rigs. I will have to play around with this. But I find those schemes to be cumbersome to use as well, though not as much so as the custom properties.
In the end there may just not be an ideal solution. They all have drawbacks. Alas. But I'm not quite willing to give up just yet.
@stinus.b: I'm not sure where that's coming from myself, but it's probably something to do with the occlusion stuff coming to funky conclusions about the hair. I'm more familiar with the animation/rigging side of things than lighting/texturing stuff :)ReplyDelete
@Cessen: Hmm... could you name some examples that I could think over. Surely there must be a way...
@All: thanks for the positive comments.
Hello, did you ever try this one?ReplyDelete
@swims: Ah cool, there's a "child" option in the toolbar in Particle Edit Mode to show child hairs too. Some interesting options there too...ReplyDelete