During the past two days I spent my time making many mistakes while rendering a 100 frame clip using Arnold as a render engine. I created a shader network for a single read node (one image input or one sequence input) that doesn't need much editing apart from altering it's exported channel layers. So far the shader network I created takes apart nearly all renderable passes from Arnold's output and provides with many different ways of editing their output of the rendered passes provided by the renderer. Examples are grading, exposure and saturation for grey-scale'd RGBA beauty renders, (in)direct diffuse/specular, normal maps, reflection, refraction, point position, subsurface scattering and deep/mid/shallow scattering. I'm still looking into how to add Z Depth as Arnold hadn't given me an image for this. Below is an image.
|Beauty, Diffuse, Normal and Object ID|
|Specular, Reflection and Refraction|
|Point Position and Subsurface Scattering (the rest of it going up)|
As you can see from the previous images I didn't bother to connect all the different outputs to the final write node as many contained information I didn't think would change the final image at all. For example, there was an almost invisible bit of information in all the scatter layers combined so I decided not connecting those. Their final result would have been negligible and it would have increased render time. Right now it took 15 minutes to render out 100 frames on my laptop at 1920x1080 resolution. That's 9 seconds per frame. Not bad at all. Below is the 4 second render this shader network managed to put out after some small tweaks.
I totally forgot to turn off the Arnold license check so I'll totally have to render it again. However, I won't be continuing this until the end of Sunday as tomorrow I'll be joining my team for the GGJ 2014. We're called 'Fat Kids are Hard to Kidnap'.