Not so long ago I worked at Lola Post, London, on another documentary hosted by Richard Hammond. Similar to the Journey to The Centre of The Planet and Bottom of The Ocean shows I worked on some time back, this entailed a heck of a lot of vfx.
The concept is that we see the constituent parts of scaled-down planets and the solar system being brought together in a large space over the Nevada desert. In order for Hammond to be able to present things at the necessary altitude, he is up at the top of a 2 mile high tower, which is obviously not real for various reasons. Nor is the desert much of the time. Or Hammond.
My input on the show was working on dust and sand particle systems. I was working on 2 sequences of shots. I will warn you now that some of this will get technical.
The first sequence shows a large swirling cloud of high-silica sand and iron. This includes a shot which was to become my baby for a month or two. It pulls out from Hammond at the top of the tower, back through the dust cloud swirling around him, then really far back so we see the entire 2km wide cloud in the context of the landscape around it. The whole shot is 30 seconds long.
The second sequence of shots shows the formation of Jupiter out of a large swirling disc of matter. Jupiter itself attracts dust inwards, which swirls as it approaches.
A few challenges presented themselves quite early on. One was creating particle systems in Softimage’s ICE that behaved correctly, especially when it came to dust orbiting Jupiter as the whole system itself swirls around the protosun. The initial swirling round the protosun was solved using a handy ICE compound that Lola have kicking about on their server, but if you use that twice in an ICE tree it is only evaluated once as it sets the velocity using an execute node, effectively overriding the new velocity value for each particle, rather than passing that out so it can be added to the previous velocity.
The solution to this was to break apart the compound. Integrating new nodes, including some out of a Move Towards Goal node, meant that I was able to make a new compound that I could proudly label Swirl Towards Goal. It sets the goal, then outputs a velocity which can be added to the velocity from the previous swirling compound higher up the tree. It even has sliders for distance falloff, swirl speed, and weight.
The most challenging aspect of this project was actually rendering. The swirling dust in each of my shots is made up of about 4 different clouds of particles. One alone has 60 million particles in it.
Enter Exocortex Fury, the fabled point renderer that was to save our bacon. Aside from one fluffy cloud pass per shot, rendered as a simple Mental Ray job on a separate lower detail cache, each cloud pass was rendered with Fury. Unlike traditional particle renderers that use CPU to render, Fury is a point renderer which can take advantage of the raw power of graphics cards. The upside is a far faster render compared to traditional methods, and done correctly it is beautiful. To speed things up further, particles which were offscreen were deleted so Fury wouldn’t consider them at all. Downsides are that it can flicker or buzz if you get the particle replication settings wrong and it has no verbose output to tell you quite how far it is through rendering. Between us dust monkeys many hours were spent waiting for Fury to do something or crash.
Adding to the complications was the scale of the main scene itself. The tower is rendered in Arnold, a renderer that works best when using one Softimage unit per metre. Unfortunately the huge scene scale caused problems elsewhere. In a couple of shots the camera is so high off the ground that mathematical rounding errors were causing the translation to wobble. Also, as particles, especially Fury-rendered ones, prefer to be in a small scene to a gigantic scene for similar mathematical reasons, they weren’t rendering correctly, if at all. The particles were in their own scenes for loading speed and memory overhead purposes, but in order to fix these issues, the whole system was 1/5 of the main scene scale and offset in such a way that it was closer to the scene origin yet would composite on top of the tower renders perfectly.
How to Build a Planet is on show in the US on Discovery’s Science channel before being shown to the UK in November.
Discovery Sci – How to Build a Planet