The Invisible, Made Tangible
Unreleased science-visualization series · 3D, simulation, rendering · 2025–26
A crystal lattice that traps CO₂, a "forever chemical" coming apart, and a particle stream filtered in real time. The series was shelved before launch. This is my cut of it, finished and rendered for this page: 15 shots, 2,108 frames.
A microscope in a dark room.
The chemistry that matters most happens at a scale no one can see, so every shot puts you inside that scale. Atoms are translucent jelly that glows from within, bonds carry warm light, and one thin plane of focus does the storytelling. Accurate first, beautiful second: real molecular geometry, real reaction order.
A crystal you can fly through.
The hero shot glides through a zeolite: a cage-like mineral whose pores can trap gas molecules.
Thirty-seven hours, then out of memory. The lattice was a Cloner that sent every strut to the renderer as unique geometry, about 3 million polygons a frame. Three cameras projected to 30–37 hours at 1080p, and the first full-size attempt ran out of GPU memory. Render instances, a headless queue and tuned sampling brought it down to one night:
| Step | Frame time, 1080p |
|---|---|
| Original setup | out of memory |
| Render instances, headless | 287 s |
| Adaptive sampling tuned | 103 s, no visible change |
| Final sampling and light trims | 57 s |
Rebuilding a lost look. The approved frosted blue survived only in a preview video; the scene had been overwritten. I rebuilt it by eye over five versions. Refraction came down from 0.88 to 0.2 so the struts stopped collapsing into one blue, three tones layered the depth, and a cool dome replaced a warm one that had been adding a pink cast. 150,000 specks of dust, visible only to the camera, bloom into bokeh and sell the scale for almost no render time.
Coming apart in the right order.
PFAS are "forever chemicals" because the carbon–fluorine bond is one of the strongest in chemistry. The original edit showed a PFOA molecule falling apart in about three seconds. I rebuilt it in Python and Cinema 4D as an ordered event, with 26 atoms, 25 bonds and real angles and bond lengths, following the pathway described in degradation research:
- the acid head leaves first, as CO₂
- the chain unzips one CF₂ unit at a time, head to tail
- each unit sheds its fluorines as fluoride
Same materials, real scale. I reused the original translucent jelly shaders exactly. Their textures are world-space, so they only read at the original micro scale, where a carbon atom is 0.46 cm across, and the whole rig was rebuilt at that size. Each departing piece moves as one rigid unit, and broken bonds shrink back into the atom that stays.
A camera that gets there first. A damped follow-cam arrives at each break about 10 frames before it happens. The first version caught the CO₂ leaving half out of frame. A render profile without motion blur or bounce light, which a black void doesn't need, took frames from 129–243 s down to 10–20 s with no visible change: 93% faster.
Fourteen looks, not 1,304.
The legacy close-ups held 1,304 materials but only 14 distinct looks. A script folded them together and repointed 2,188 tags, and the 14 were remapped onto the PFAS jelly palette. With the shared dust, the series reads as one world. Each camera was cut to its strongest five seconds, picked from contact strips of the whole move.
Focus as a design choice. The original macro depth of field hid the chemistry. Halving the blur kept the microscope feel and made the structure readable; quartering it tipped the shots into a diagram.
A simulation that only plays forward.
A stream of particles flows into a filter. The red ones, captured CO₂, are caught in a wall that swells into a sphere; the blue ones pass through in rippling rings.
The caches were gone. The simulation's playback files had been lost, and a search of every drive came up empty. The original rig was still intact, so I re-simulated it. But it only plays forward: jumping straight to frame 100 gave 224 particles instead of 22,334, so any render that started mid-shot or resumed after a crash would have been wrong. I cached it to a 4.6 GB Alembic file and checked four frames against the live simulation: the particle counts matched exactly.
Picking the shots. The scene held 11 cameras. I rendered each as a strip in the final look, picked five, and reviewed a low-res animatic for motion before rendering a single full-size frame.
Clay underneath.
Finals rendered from the command line with the interface closed, 3–10× faster than the in-app queue, on one laptop GPU that sat at 84–87 °C and throttled hard. A faster setting was used only if it matched the approved look, and every finished sequence was scanned for camera jolts and flicker. For the breakdowns, a script swaps every material for one matte grey and hides the dust, so the same cameras and animation render as pure form.
Stills from the 4K renders.
Who made it.
- 3D direction, look development, simulation, lighting, rendering, finishing
- Mason Barry, Barry Creative
- Tools
- Cinema 4D 2026, Redshift, MoGraph, C4D particles, Python, ffmpeg