Claustrophobia

Traveling the incredible distance of 20.000 miles - almost once around the globe, which has an equatorial circumfence of 21600 nautical miles by definition - submerged deep below the sea in a prototype submarine would probably have been a nerve-wrecking experience for Jules Verne's heroes, especially with the prospect of being trapped in this vessel (how adequate a word!) for their remaining lifetime. They must have felt like goldfish in a bowl - a vessel which, fortunately, in most countries is nowadays illegal to use as a fish tank. Like with deep sea diving equipment of Jules Vernes' times, the main issue of the goldfish bowl wasn't even the claustrophobically small dimensions, but the problem of providing an adequate amount of oxygen. Unfortunately for countless goldfish, they did not get a chance to mutiny like Jules Verne's protagonists.

This scene was heavily inspired by Peter Hertel's awesome "Trapped Wilderness" (January 2008 stills competition's 3rd place).

The idea developed a few days after I stumbled across the TINA CHeP pages, and I pulled this one off within 3 days.

The first day was entirely dedicated to creating the diver's helmet, which is a set of meshes I did from scratch in Wings 3D. For all you out there just starting on subdivision surface modelling - go get it! It's incredibly easy to learn despite the quirks it does have. Took me the whole day though, to get all the details right, choose proper materials, and set up the basics of the remaining scene.

Second day's work was focused on creating the fish; I found a good collection of free fish meshes on http://toucan.web.infoseek.co.jp/3DCG/3ds/FishModelsE.html, but since I definitely wanted a simple common goldfish and they just had a decorative one, I decided to instead go for the "common carp" mesh (after all it is a close relative of the goldfish), which I tweaked a bit (again using Wings 3D) to better match the common goldfish shape; the texture map I re-coloured and added transparency to the fins using Photoshop 6.0 (hey, can you believe this - I've been using the software for years now, and it took me till now to discover the "Export Transparent Image" wizard in the help menu). I guess I spent 2 or 3 hours on that.

On the same day I also settled my mind about the total layout of the scene. I went for a plain white room lit only by radiosity from an overly bright, purely white sky sphere (White * 1.8), using light groups to add extra lighting to the helmet from a dim point light source, and even more light from another light source to the fish which would have turned out too dim otherwise. The helmet and fish got placed off-center and the camera moved back to add an emphasized sense of isolation.

The third day was almost entirely busy taming the quirks of radiosity. To begin with, POV-ray developed a nasty habit of crashing as soon as I moved to high-quality radiosity settings - to an extent that I wondered how on earth such a bug could have gone unnoticed by the POV-ray community until now. Having initially suspected my meshes to be bogus somehow, it took me at least an hour to find out at last that histograms - which I had happened to have activated just for fun - really don't go well with radiosity. When I finally got to get high-quality radiosity settings running, there came the challenge of getting the settings right. Turned out that the high-quality settings from the radiosity tutorial didn't work at all for my scene (just as the tutorial promised :)), and the best settings were much closer to the default than I initially expected.

-- update --

I just gave it another shot, and found that for some reason it does make a significant difference to render the image once, saving radiosity data, and then rendering again using the saved data. The additional pass took 10 minutes. I also used a higher radiosity sample count, resulting in a first-pass rendering time of 8 hours 33 minutes, but since there was not much of a difference to my initial submission, I guess that didn't add much to the quality of the final output.