Hi guys,
Well, I didn’t really get a chance to do a proper write-up (which I still intend to do) but in the meantime, the following should explain some of your questions.
Basically, I had to do a great deal of testing with Vray for character work as we have recently finished a 6 minute fully computer generated short film featuring a character and an indoor environment. To do a full GI render in one layer, the rendertimes exceeded 2 hours/frame. Even with Photon mapping I wasn’t able to get things to speed up AND retain a good image quality. It was this that led me to figure out a faster method of rendering such projects.
The bottom line is that characters must be rendered using a single frame GI calculation. This is because characters move and every frame of the GI lighting is different and therefore has to be re-calculated. This is different from a camera move fly-by in which the background does not move and one can therefore pre-render the irradiance map at every n’th frame and then render the sequence over a network.
Now because characters must be rendered with single frame GI, the fastest method for doing this is to exclude all other geometry from the scene and render ONLY the character. This is nowhere near as accurate as if you rendered the character with the room “Not visible to camera” for instance, where the character would therefore interact with the environment lighting. However, this is also an extremely fast render. The problem is, if you render the character by itself, on it’s own layer, how do you make it interact with the environment lighting?
One solution which I find works well, but is not necessarily accurate is to use the rendered enviroment as a lightmap. Some of you have asked what this means. Lets take a scene example in which you have a character in a blue room with red cast lights. First, hide the character from the room and render the room as a layer of it’s own. For this, you would render it with all the bells and whistles (GI, Photon mapping, whatever). If there is a camera move, you can prerender the irrandiance map every nth frame and then render the entire sequence over a network.
You now have your background layer done.
Now you want to render your character. First get rid of all the room geometry in the scene (hide it like you did with the character) and unhide your character. Note, that you will keep exactly the same lighting set-up you used for your background (ie. the red lights) as you obviously want the character to be lit from the same direction, intensity, color, etc that the room was lit with.
Next, because the room is hidden, the room environment lighting will not be cast onto the character. You can fix this by using the rendered image(s) of the room layer and dropping it into the “Environment” slot (for GI) in the Vray render menu as you would with an HDRI. As with any environment map, make sure it is set to ‘enviroment’, not ‘texture’. Now turn on GI and do a render.
You notice that the character is now lit with the background colors, as well as the scene lights. The first problem that often arises from this is that the combination of the scene lights with the background map GI is too strong and your character gets blown out. This will require minor light tweaking such as lowering the intensity of the main lights, as well as the GI Map.
The second problem you will notice is that the character is lit with the background map as if the map was ‘projected’ directly on top of the character. This means that if you had a red square in the map, the red square would be projected onto the character and this is clearly visible, especially when the character moves throughout the environment. There is a fix to this.
Blur the crap out of the background. The reason is because you are not interested in the background details - you only want the background colors to be used for your character. NOTE: Turning up the blur in the material DOES NOT blur the environment map even though it shows it blurred in the material editor! The right way to do this is to bring the map into Photoshop and put a Gaussian blur of 70+.
There is no specific way to do all this. It changes depending on the scene. For example, if you have a camera move that moves from one end of a room which is blue to the other end of the room which is red, you would use the rendered sequence of the background as a light map because it is important to reflect the change in the background on your character.
By doing it this way, you can have blazing fast GI renders of your characters since there is hardly any other geometry in the scene complicating the GI solution. To give you an example, I usually use a GI min/max resolution of -2, 0 for final production renders with an aliasing of 0,2. With these high settings, the GI map calculation often takes no more than 5-10 minutes. I usually allow myself a maxium of 30mins to 1 hour for final frames with full GI, reflections and motion blur.
There are a whole bunch of other things you must keep in mind when rendering over a network such as rendering your irrandiance map on one computer only and then making it accessible by all render nodes.
Other things which help integrate your character into the enviroment is to use your pre-rendered enviroment map as a reflection map as well. Once again, remember to turn it to ‘environment’ not ‘texture’ and make sure it is on ‘spherical’.
There are also occasional exceptions to the above case. For instace in Robaxacet:
http://www.richardrosenman.com/whitehall.htm
There is a scene in which the characters come running out of the house. The inside of the house is lit with warm yellow lights and the outside is lit with cool blue lights. It would have been impossible to create a convincing effect of the change of light using the pre-rendered background map as a lightmap because basically it would light the characters on the top of the screen warm and those on the bottom of the screen cool. In addition, the fact that warm light floods the cool lights when the door is opened only complicates the scene. In this instance, it was necessary to build the front of the house with dividing walls so that the interior which was lit with warm Vray lights would contain the light and separate it from the exterior cool Vray lights UNTIL the door was opened. Naturally, having included geometry in the scene that is not visible to the camera but still contributes to the GI solution slows down rendering quite a bit.
Ok, I think this post is long enough. Hopefully I’ve clarified a few things for you guys.
To sum it up, the above method is not necessarily physically accurate but it looks good and it renders damn fast. And in the end, that’s all that matters. 
I still intend to do a proper write-up regarding this which I should have ready within a week or two. I’ll post it as soon as it’s ready.
-Richard
