Here’s some of my observations with them in the context of artists animating a referenced rig in production:

• StoredExpressions doesn’t always store expressions. They actually show up in StoredPositions almost every time you touch something with an expression. They often also break / stop evaluating when this happens. (They often get stored even when you disable storing of expressions in your Delta, but it’s hard to repro.)
   
• StoredPositions remembers transforms by the mere act of selecting and having translate/rotate/scale tool active. While not exactly a bug, it easily causes problems if you edit centers later on, because the transforms get remembered and so it causes a wacky offset.
  Stuff can exist in both StoredPositions and StoredFCurves but it shouldn’t. If it’s keyed/animated, keys take priority, which means you got redundant junk in StoredPositions. XSI won’t remove the StoredPositions entry when setting a key (which makes a StoredFCurves entry.)
   
• StoredPositions will (from time to time) remember changes to a referenced material’s parameters. This combines with the annoyance of updating a material later and having old values mysteriously enforced on it because they were in StoredPositions and shared the same names. (Maybe a better name is “StoredValues”, since it stores more than positions.)
   
• Generally rigs don’t come preanimated so there’s no point for “StoredRemoveAnimations” to exist. It gets created whenever you remove keys from something in the rig.
   
• StoredConstraints remembers constraint relations by explicit naming. Sometimes these don’t update when the objects name changes, and sometimes they do.
  So say if you constrain an arm of the referenced character rig to a null you just made, THEN rename the null, there’s a very high chance when you save and reopen the constraint won’t be connected, since it will be expecting the original null name. The good news is you can open the Delta in an Explorer, doubleclick the StoredConstraints subproperty and rename the object names in each row to whatever are the correct ones.
   
• Whenever you edit contents of a Delta by hand and want to see the updated result, you must rightclick on the referenced model in an Explorer and “Update Referenced Model.”
   
• You need a Delta per model… and in the case of having a rig with one or more models inside it, this rule still applies.
   
• Just because a Delta sits below a model that doesn’t necessarily mean that is the model it’s working on. It’s actually working on whatever its “Target” parameter specifies.
   
• If “Record” isn’t ON (in the Delta’s PPG) before you make changes, they don’t get saved. End of story. — Turning it on after doing changes won’t save your ass. (Good thing it’s on by default, eh?)
   
• If you have a convoluted/complex rig that crashes when you do “Update Reference Model” or when simply offloading, but works fine otherwise… if you need to update the refmodel path, you can do this without opening your scene by editing the relevant scntoc file and then when you open your scene it’ll be path you wanted.
   
• When you duplicate a material from a referenced model, it duplicates fine EXCEPT for the image clips/sources, which stay referenced and when you reopen the scene, they will have dissapeared.
   
• Using Animation Layers with referenced models is not to be trusted. [*update*: Steven Caron from Blur Studios has noted that as of Soft 2010 SP1, animation layers work fine with referenced models. Beware if you are using a previous version.]
  If you use them, always open the scene you just saved in a new XSI without closing the first one so you can check that when it opens all the animation is still there and works properly. (I haven’t confirmed, but I believe I read on the mailing list that you can collapse animation layers before you save and then it’s okay.)
   
• Someone on the Softimage mailing list mentioned once that only the first Delta for a particular model remembers Mixer stuff like actionclips and such, regardless of what you set. Good to bare in mind.

At work I made two relatively simple scripts that help us with refmodel woes. I’d like to share them with you:

This really simple script imports a refmodel, make a fresh empty delta for it (and extra Deltas for any models in it) and sets these defaults automatically:

• Static Pos and Other Parameters ON
• FCurves ON
• Expressions OFF
• Constraints ON
• Animation Mixer ON
• Group Relations ON
  (this stores partitions relationships so it’s very important!)
• Model Instances OFF
  (could be left ON if you need it)
• External Connections ON
  (this remembers stuff like constraining something in the refmodel to an external object)
• New Properties OFF
• Clusters OFF
• Cluster Properties OFF
• Operators OFF

then in Mixer Modifications:

• Animation ON
• Shapes OFF
• Audio ON
• Cache ON

These are the ideal settings for using a referenced model to move/animate it, put it in partitions/groups/layers and render.

Download the AF_ImportRefModelForAnimation.py Python script here. Make it into a button or incorporate it into your workflow tools if you have any.

“Clean Delta”:
It deletes stuff from a Delta that shouldn’t be there if you just wanted to animate and render a referenced rig or model. See the code for more specifics.

Download the AF_DeltaCleaner.py Python script here and run it from a button with refmodels selected. I’ve left plenty of code comments to make sense of what it does. I urge you to read them.

I hope these scripts assist you in having a more pleasant experience with referenced models. They’re actually quite wonderful after you learn their quirks.

If anyone has anything else to say about deltas, modifications, suggestions, code improvements, constructive critique, cheese and monkeys, etc. please leave a comment. Thank you.

The following Python script iterates through your selected enveloped objects, creates SymmetryMaps if one doesn’t exist, and then goes through their envelope deformers and appends them to the SymmetryMappingTemplate for that model based on the naming convention of prefixing “L_” for left-side and “R_” for right-side objects.

It assumes the following: (and maybe in the future I’ll make the code smarter…)
- Your selected objects all have envelopes.
- Your left-side objects begin with “L_” and the right-side with “R_”.
- All your selected objects belong to the same Model.
- That you don’t have more than one SymmetryMappingTemplate under your model. (It will append to the first one found under the model, and create one if one isn’t found.) I recommend not having one to start with.

Make yourself a button with this:

xsi = Application
lm = xsi.LogMessage
from win32com.client import constants as c
import win32com.client
 
objects = win32com.client.Dispatch( "XSI.Collection" )
objects.AddItems(xsi.Selection)
 
# Iterate through selection
for envObj in objects:
	lm( "Iterating through: "+envObj.FullName, c.siVerbose )
 
	# Create a SymmetryMappingTemplate if needed.
	oSymmetryMapTemplate = xsi.Dictionary.GetObject( envObj.Model.Name+".SymmetryMappingTemplate", False )
	if not oSymmetryMapTemplate:
		oSymmetryMapTemplate = xsi.CreateSymmetryMappingTemplate(envObj, False, 0, False)
		lm( "Symmetry Mapping Template not found; created one.", c.siVerbose )
 
 
 
	# Create a SymmetryMap if it doesn't exist.
	sMapCls = envObj.ActivePrimitive.Geometry.Clusters("SymmetryMapCls")
 
	found = False
	if sMapCls:
		sMap_props = sMapCls.LocalProperties
		for eachProp in sMap_props:
			if eachProp.Type == "map" and not found:
				sMap = eachProp
				found = True
				lm( "Found SymmetryMap: "+sMap.FullName, c.siVerbose )
 
	if not found:
		lm( "SymmetryMap not found; created one.", c.siVerbose )
		#sMap = xsi.CreateSymmetryMap("", envObj, "Symmetry_Map")(0)
		sMap = xsi.CreateSymmetryMap("", envObj, "", "")(0)
 
 
 
	# Add envelope deformers to template by name:
	i = xsi.GetNumMappingRules( oSymmetryMapTemplate ) + 1
	oppositeNaming = {"L_":"R_", "R_":"L_"}
	envelope = envObj.Envelopes(0)
	deformers = envelope.Deformers
	for eachDef in deformers:
		name_A = eachDef.Name
 
		if name_A[:2] in oppositeNaming.keys():
			name_B = name_A.replace(name_A[:2], oppositeNaming.get(name_A[:2]), 1)
 
			xsi.AddMappingRule( oSymmetryMapTemplate, name_A, name_B, i)
			lm( "Adding Mapping Rule #"+str(i)+": "+name_A+" <--> "+name_B, c.siVerbose )
			i += 1
 
xsi.SelectObj(objects)

I might update the code in the future to support multiple models and check if an entry in the template has been added already, but for now this above will have to do.

Hope it’s useful to somebody!

With this Rendertree compound you have AO in darkness and that’s it. Simple!

Now….. can you spot the differences between these images below?

Diffuse pass, boring looking, no AO:

BAD — Sloppy AO that’s been applied everywhere: (notice how dirty it looks towards the center light)

GOOD — Shadow-aware AO, only exists where it should… in darkness:

Credit where credit is due; this method was originally pointed out by Guy Rabiller many years ago, back when our software was called a 3 letter word.

However, the way it’s designed relies on the existance of the addon to evaluate. Without it, your rig is broken. I’ve made a version which is self-contained, so when you export your rig you don’t need to tell people to install any addon.

Select your curve(s) and run this code from either a button or the Script Editor. Here’s the JavaScript:

// Self-Contained CurveLength Op -- http://darkvertex.com/
 
function CurveLengthOp_Update( In_UpdateContext, Out, Incrvlist ) {
	Out.Value = Incrvlist.Value.Geometry.Curves(0).Length;
	return true;
}
 
function ApplyCurveLengthProp() {
	oEnum = new Enumerator( Application.Selection ) ;
	for (;!oEnum.atEnd();oEnum.moveNext() )
	{
		var sel = oEnum.item();
		if (sel.Type == "crvlist") {
			var oProp = sel.AddProperty( "Custom_parameter_list", false, "CurveLength" );
			var p = oProp.AddParameter2("CurveLength",siDouble,0,-8000,8000,-8000,8000,siClassifUnknown,siPersistable | siAnimatable, "Length");
			var crvList = sel.ActivePrimitive;
			var newOp = AddScriptedOp( p, CurveLengthOp_Update.toString(), crvList, "CurveLengthOp", "JScript" );
 
			newOp.Debug = 0;
			newOp.AlwaysEvaluate = true;
			newOp.Connect();
 
			LogMessage("Added CurveLengthOp to: "+sel.FullName, siVerbose);
		}
	}
}
 
ApplyCurveLengthProp();

Fiddling with the ExtrudeOp options

 
Try this yourself:

1. Get a sphere. (My screenshot example was a half-sphere.)
2. Select all polys. (Press Y to enter poly mode then Ctrl+A to select all.)
3. Press Ctrl+D to Extrude.
4. In an Explorer open your object’s operator stack, find the ExtrudeOp and open its PPG.
5. Turn off Merge, then fiddle with Length, Inset Amount and Subdivs.
6. Then in the Transform tab, play with the sliders there.
   You can get some very interesting (and often abstract) results!

[CLICK TO SEE VIDEO]

BREAKDOWN:
1. Set an OK center position with "Move Center to Bounding Box" / "Move Center to Vertices".
2. Enter "Center" mode. (top right)
3. Right-click on "Ref" button (below Local) and choose "Pick Polygon/Edge/Point Reference".
Click an object to select it and highlight the desired components.
4. While still in both "Center" and "Ref" modes, reset rotation to 0,0,0 then go back to "Object" mode (top right), go "Local" mode and reset rotation to 0,0,0 again. — That’s it!

FORUM THREADS: [xsibase] [xsiforum.com]

[CLICK TO SEE DEMO VIDEO]

FORUM THREADS: [xsibase] [xsiforum.com] [softimage.ru]

WHAT IT DOES:
Inserts a <script> tag into every htm file in the docs folder so that when you click on any paragraph, it will be automagically translated to your chosen native language. (The video above shows a Spanish demo.) Language is set when you run the patching script.

I was responsible for creating the ICE compound to position phone instances on a curve and simulate the domino effect.

** Download Example Scene (XSI 6.02+) ** — Raw Rotation & with post-blur [QT MP4]

Why?

The fast skinshader (mentalray) material that comes with XSI off the bat is pretty awesome, but it has one downfall which is the lack of a transparency input.

How?

A Refraction node combined with Binary Alchemy’s wonderful SSS shader, from their (freeware) "Essentials" shader collection, available for download on their site.

BA’s SSS shader doesn’t do transparency or refractions off the bat, and it won’t let you plug in a material, so that’s not an option. However, a Refraction node (or Transparency node, though that would look wrong) can be plugged into its "Replace Color" input no problem. Plug your transparency map to the weight input and you’re set.

The final rendertree looks like the following:

But why would you need skin to be transparent? Humans aren’t transparent, you crazy nut!

True, however… what about fish? Certain fish can be almost completely transparent and most are at least partially so.
Not only that, but in the areas that they are, they refract light, too. Behold some evidence below:

What is the refractive index of transparent skin?

That’s a good question…. and I’ve no idea. (If anyone knows, get in touch.) I’d guess it’s something around that of some sort of plastic or silicone.
(In my example scenefile I used refraction of 1.2 just to see if it was working and by no means whatsoever is that the correct refractive index for skin.)

Ok… it’s transparent, but shouldn’t it be a bit blurry too?

Absolutely! And that’s where the magic of post-production can kick in. One way to do it is to have a pass of your transparency as black on white, and use that to selectively blur the transparent areas to give the illusion of blurry refraction. [Another way, potentially slower to render, is to use an Architectural node instead of the Refraction one to get blurry refractions off the bat. -- Thanks ghonma for the tip.]

I’ve provided my simple approach to doing the effect in post using XSI’s own FXtree:

(Click on the picture to the right to download a QT .mov of the animated rotation after post-production.)

If you see any mistakes or feel the need to correct any of the methods described above, let me know so I can make this document more accurate.

[SEE VIDEO]

FORUM THREADS: [xsibase] [cgtalk] [xsiforum.com] [indocg.com] [xsiforum.de]

NOTE:
Sometimes, enabling “Sample Displacement Before Rendering” under the Optimization tab in the Renderer Options may drastically speed up your render even more.
Bare in mind disabling it means mentalray doesn’t get accurate bounding boxes for the displaced geo, which means BSP calculations can be affected.
This will not affect render quality though, only overall render times.

*UPDATE* [03/01/10] This tutorial was made in XSI 6 and you should know that the “sib_interp_linear” node is the “ChangeRange” node in present versions.