The film lasts almost five minutes and is comprised of 40 scenes. It was rendered in 24 FPS in lossless HDR format, at 2K Scope size, with Aspect Ratio of 1 : 2.39, having a depth of 2^96 color shades, that is 79,228,162,514,264,337,593,543,950,336.
Because there is no standardization in how films are made, every film requires custom technology. A “full” arsenal is often inadequate. Using a bouquet of tools – probably a dozen – plus proprietary technology was a more reliable solution. Panayiotis developed a solid pipeline for data synchronization. Although it involved some extra time cost, there was no data loss. So, his film is software independent.
The computation load was tremendous and would have taken ~12000 hours on a single CPU of modern technology to process. Proportionally, a full-length feature film would have taken 48 years on a single CPU to compute. Needs and technology are in constant battle. Even attempting to repeat such endeavor at the peak level in the future, the computation time would not vary much.
The load was distributed on the farm of RenderStreet in 2 phases, preview and final. The respective rendering times were as follows:
. 1K Draft proxy preview rendered locally – 3 hours, Raw Size 1.7GB
. 1K Full preview rendered remotely – 1 week, Raw Size 8GB
. 2K Brutal final render, rendered remotely – 40 days, Raw Size 34GB
The film was “developed” in a Virtual Darkroom (baths into chemicals) in order to simulate stock film tonality. This, provided highlights roll-off and subtle Selwyn granularity found in high-end movie cameras. Neat!
“Lightgambler”
When trying to interpret infinity, there is always noise due to finite number of samples. In visual terms, noise shows up as flickering on moving imagery (baked frozen walkthroughs excluded). You can’t eliminate noise, it’s a utopia. Eliminating remaining noise would require exponential calculations. It does not worth. The objective goal is to reach accuracy by weakening the noise not to disturb, without taking countless samples. Guessing is inaccurate, blurring digests details, thus both forbidden.
It was in early 90s when I started experimenting with light in digital terms with focus on Global Illumination. A few years later, my experiments drove me to some awesome results. I coded my invention and named it “Lightgambler” due to its statistical nature. It was a technology ahead of its time. I thought “How about making a film using it?”. “Lightgambler” was the “raison d’etre” of my film and apparently the name of the film’s website.
It was in the last 90s, when I was showing online, some images to peers. A world reputed Spanish 3D Guru (M.F), was asking me how did I reach these results since there was no such technology available. Disclosing was devaluing. I let him wonder. My goal was to utilize my invention for my needs rather than to dethrone anybody. Technology evolves very fast and most probably there are already better solutions. In any case, it worked for my needs.
Lightgambler took shape in 2000 and was smoothed in time. Apparently, it is not a parthenogenesis. There is knowledge and reason behind it, some science and passionate treasure hunting. Rather than dealing with static imagery, Lightgambler was built to deal with moving imagery where light inconsistency is more profound especially in light inaccessible areas. In the case of my film, Lightgambler was proved very effective in computing illumination in indoor scenes exposed to sunlight, which are cases most CG films avoid or cheat/compromise. In a particular shot where Thespis leaves the house, while closing the door, the light behind Venus de Milo is so realistic! Lightgambler contributes to consistent, cleaner results, faster. It still surprises me.
A code driven multilayer kinematics rig (armature) that allows the protagonist to act in a progressive fashion, move faithfully without breaking the anatomy, interact smoothly with the environment, perform hand gestures and facial expressions. It’s a hybrid system that allows both, automated and manual control at any time, making character animation a joy. (2000)
Skin shader with subsurface scattering. (2022)
Hair shader. (2023)
The camera is attached on a crane rig for accurate procedural movement. It allows 3 modes of operation (locked, loose, targeted) with smooth transition between the three modes. Camera Parameters are controlled within the rig.
The Honda XR 500R
The particular model is a dirt bike that was introduced in 1983. It was Panayiotis favorite bike at his adolescence. He used to draw it on paper so he was familiar with its anatomy. The bike rig allows Thespis to ride it with front and rear suspension, accurate steering, wheels and chain rotation. Many tiny details were omitted for practical reasons.
Other details
Venus de Milo is a 3d scan by Sketchfab. All other models were made from scratch based on references. Apart Honda XR500R, the Beatle Volkswagen, the Singer Sewing machine and the Bedouin matches, all other brands were fictitious and mocked. There are many details that cannot digested by viewing the film once.
All the animations were made from scratch. Cloths displacement, liquids, pyrotechnics, ground fog and other volumetrics were computed procedurally. No templates, no motion capture, no artificial intelligence, no rotoscoping were used. Automated solutions encourage laziness. It’s admitting defeat. They generate mass similar results, that alienate creativity and eliminate the creator’s stigma from the production.
“There is no perfection. The film already contains several flaws. Some of them are forgivable, some of them are due to software bugs that I couldn’t do anything. As long as the film messages are received, I am pleased. If I had the chance to redo the film,
I would have avoided them, but … I am learning from my mistakes.”
Films made with real or artificial means have the same root, the need of filmakers to express and communicate. Both courses are very similar, converging more and more. Nowadays the separation line is blurry and eventually will become indistinguishable.
Still from James Cameron “Avatar” film, 2009.