Global Illumination and Final Gathering in Softimage XSI 3.5.1
-part 1-

© 2004 Rob Wuijster



Introduction:

Besides the standard raytracing and lighting capabilities of Mental Ray, there are also some fairly new lighting techniques you can add to your scene. One is called Global Illumination, the other one is called Final Gathering.
(A third would be Caustics, but that would be another tutorial....)

Simply put the first technique incorperates the "bounced light energy" from object to object, and taking the color of the first object over to the next one. Like a red ball on a white surface, the ball will "spill" it's red color onto the white surface beneath.
The second technique allows you to use ANY object in your scene to become a light source. This technique is an excellent addition to GI or HDRI rendering.

This tutorial is greatly based on a Maya tutorial I found on the web. It's only fair to give the author of that tutorial most of the credit, because I mostly translated it to XSI lingo and workflow. Yes, he knows.... ;)
First steps, creating the geometry:

Start up XSI and create a Cube with "Get>Primitive>Polygon Mesh>Cube". Change the subdivs to 5, 5 and 10 and scale the cube in the Z to a value of 2. Freeze the scaling, so the extrusions will be uniform later on.



To get a interesting effect with the GI rendering, we're gonna make our cube a bit more dense. Select all polygons of the cube, except the "backwall" faces. The easiest way to do this is to Rectangle select (Y) the entire cube, then deselect the "backwall" faces again with the middle mouse button in the Top view.

Now duplicate the selected faces by pressing Ctrl+D or by "Modify>Poly.Mesh>Extrude along Axis" or by Alt+Right Mouse button "Extrude along Axis". All this creates a "Extrude Op" under the Cube geometry.
In the Extrude Op set the Extrusion length to 0, and be sure that the extrusion is set to the Y axis. Also de-check "Merge", so the faces will extrude on their own.
Switch to the "Transform" tab, and set the X/Z scaling values to about 0,8.
Be sure to have all the translation values set to 0.

With the faces still selected, extrude the faces again in the same way. A second "extrude Op" is created. Set the Extrusion length here to 1, again be sure that the extrusion is set to the Y axis. Again de-check "Merge". Switch to the "Transform" tab, and be sure to have all the translation values set to 0.



Note:
You could extrude "the other way around". Keeping the extrusion lenght at a value of zero, you can use the Y transslation to extrude the faces. Be aware that using extrusion length AND translation will multiply the two values.

For the final part of modelling select the inner set of 9 "backwall" faces and delete them, creating a square hole in the cube. Now (unhide the camera and) position it inside the cube, somewhere in the corner, looking back at the opening you just created. To get a better view, open up the camera PPG and change the FOV to about 70 degrees.



Note:
You should get yourself in the habit of saving camera views, by working with the "Memocam" buttons. Not only you can switch quiclky between camera angles, but also FOV values and more are stored with the camera angles. Also every view in every port has it's own settings.
Setting up lighting:

Delete the default light from the scene (I always do..) and get a spotlight by "Get>Light>Spot" and position it in front of the cube, pointing it inwards by moving the interest towards the front of the cubewall. Keep all the Y values the same so it's horizontal. Adjust the cone angle and such with the "B" shortcut, or open the PPG and work from there.



Select the light and open the PPG. Select the "soft_light" tab and turn on shadows, and set the Umbra to 0 to get dark shadows.To get a more realistic light behavior, you can turn on "light Falloff" in "Light Attenuation". If you do you have to set the Start and End Falloff values and play with the Mode setting. I will stick with the default setting here.

The're a couple more things that we have to do.

First assign a Lambert shader to the cube. Select the cube and do "Get>Material>Lambert", change the color to a red-ish color, like the XSI colors ;)
Secondly we have to turn on the GI for the object itself. Select the cube and in the "Visibility" node turn on "Global Illumination Transmitters" and "Global Illumination Receivers" for the cube, so Mental Ray will take the cube into effect with the GI rendering.
Thirdly adjust the Scene Ambience to black, instead of the default darkgrey color. This will help the color of the shadows a lot. Draw a renderregion to see what we got so far. Nothing special yet, because in the renderregion the GI is not yet taken into account when rendering.



So let's setup the global illumination.
First select the spot and in the PPG select the "light" tab, then goto the "photon" tab and turn on Global Illumination. This will tell Mental Ray that the spot emits photons. We'll keep the default values here for now.

When we keep working with the Renderregion we have to set this up for Global Illumination as well. In the render panel go "Render>Region>Options...", and in the "Photon" tab turn on Global Illumination. If you have set the renderregion to autoupdate, the GI will render immediately.
You'll have to do this for the final render settings as well, or your renderpreview will look a lot different than your final renders ;)

As you can see, Global Illumination adds a lot of additional lighting to the scene, so it's wise to adjust the strenght of the spot most of the time.

The other problem is that the quality isn't very good. We have to fine tune the Global Illumination so that it looks a lot better. So how can we tune the global illumination? You've seen the "accuracy" and "radius" values in the render region, or the "intensity" settings in the spot PPG.

Options in the spot attributes are used to control the amount of emitted photons in the scene and the brightness/intensity of the Global Illumination effect. The options in the renderregion are used to fine tune the effect, also based on radius (XSI units/pixels). This will be explained in the next part of the tutorial.

Note:
When working with more complex scenes, it's best to get use the "group" functionallity in XSI. It's a lot easier to assign rendersettings this way, or just add or remove objects to and from these groups. For instance a table with several objects on it could be split up in two groups. One group would contain the table and be assigned the receiver property. The other group would contain all the objects on the table and be assigned the transmitter property. A newly created object on the table can easily be added to this group, inheriting all rendering properties at once. There's no need for settings these properties over and over again per object this way.