RE: Reference model limitations?

["Andre DeAngelis" <andre.deangelis(at)ubisoft.com>]

Sorry Gentlemen, but with all due respects, this is ideological rigidity on your part.

 

It’s like saying that a rendering pipeline is only valid if it can achieve the desired result in a single pass.  Think about it gentlemen, what good is a ref model when you need to need to take stuff into Maya?

 

Pipelines are about achieving the end result with the minimum amount of pain, so the ultimate aim is efficiency, which in turn implies breaking your assets down into the smallest and most easily recoverable granules.  Reference models themselves represent only one type of referenced data/asset, and being a complex chunk of data, are more prone to corruption than more granular components such as UV data, envelope data, etc.  There reason point caching and ASCII/XML have become popular is because they not only provides a simple interchange format between 3D apps, but because they provide great flexibility when sharing data between animation, FX and lighting.  Why send a fully rigged character with constraints/SCOPS etc to lighting, when they have no need for all that junk?  All you are doing is introducing inefficiencies into the most critical part of your pipeline, which is lighting.

 

Animators don’t generally care bout Syflex operators and particle emitters, so why should they be forced to deal with them?   Texturing artists don’t care about rigs so why should they be forced them to deal with them?  All they want is a powerful and efficient rig to get the job done with a minimum of fuss.  Brad Gabe could probably explain this further, but he has mentioned the trend towards converting assets towards smaller, easily reproducible components.   After all, if you are going to reply on ref models as a crutch, you’d better pay that you don’t have a mixed pipeline for modelling and FX (which is usually the case).  What do you say to those poor studios who use Maya, Houdini or in our case XSI/Maya/Max?

 

Why send an entirely rigged character to lighting when all they really need is the point cache file?  Is not a point cache file just as valid and powerful as a referenced piece of data as an entire reference model?

 

I am the first to admit that there is work still to be done on Deltas and making Ref Models more robust, but it pays to use them wisely.  Reference models are certainly valuable, but even if they were bullet proof, they are not the be all and end all of pipelines.

 

Andre

 

 

---

From: owner-xsi(at)Softimage.COM [mailto:owner-xsi(at)Softimage.COM] On Behalf Of Jeffrey Dates
Sent: July 26, 2007 11:57 AM
To: XSI(at)Softimage.COM
Subject: Re: Reference model limitations?

 

>...so far I am forced to admit that there is no workaround to my problem except committing to local which is something I translate into pipeline failure.

 I couldn't agree with this sentiment more.

It sounds like you're doing the most with Ref models that can be done in a pipeline. Use them as reference models as long as possible to facilitate changes, and localize them when left with no other choice. ( usually during lighting and FX )  This is how the character work I've done usually goes.

Often times, I'd keep a scene file that maintains the Reference, and a corresponding scene that is localized for render.  If character changes do get propagated, it's just a matter of determining which is less work... going back to the referenced scene and redoing my localized work, or applying the changes to my local scene.  Neither is ideal.

The other option is to either manually, or via scripting, switch out local models and replace animation via clips or something similar.   I've just found this prone to user error. ( or scripting errors ), but find that a majority of the studios just simply work this way.

As for your specific issue of collision objects on referenced cloth....  Basically, cloth sim/collission should be the last step on your pipeline before render... 
You might try localizing just the cloth as a seperate model and constraining it to your ref. model.  Or localizing it, colliding it, and baking it with a point cache.  But in the end, just going local is the least headache... 

There is very little discussion about  Reference models  in an XSI pipeline, and find that very unfortunate.  Modular, Referenced Data is so powerful, and ideal for non-linear production, it astounds me it's not the 'norm', or more heavily implemented in the software.  

Hopefully someone else is pushing this workflow in areas I'm unfamiliar with...    I'd sure welcome it!

--

Jeffrey Dates
www.kungfukoi.com
214.280.4353