Two of us, one hall, and a rig that reads 187 000 frames a second. Glass, cast concrete, porcelain, river ice — we put it in the blast box, break it on purpose, and hand you the four milliseconds nobody has ever seen.
Concrete fails slowly enough to read at two and a half thousand. The crack front travels at roughly 1 400 m/s, so a 300 mm cube is over in 0.2 ms — but the dust cloud that sells the shot takes another two seconds, and that is the part people watch twice.
The interlayer is the whole story. Glass goes at 1 500 m/s and the shards would be gone in three frames at 2 500 — at fifty-four thousand you can watch the vinyl stretch, hold, and finally let go a full millisecond after the glass itself has stopped being glass.
Our fastest material and our most expensive frame. At 187 000 the sensor drops to 256 × 128 pixels, the window of recording is 0.9 seconds, and we burn 92 kW of light to get an exposure at all. One plate, one chance, four people in the hall holding their breath.
A shoot day is nine hours of rigging around an event that is shorter than a wingbeat. This is what those four milliseconds are made of.
The head is already recording and throwing frames away. It holds 11.4 seconds of 11 000 fps in RAM and overwrites itself until something tells it to stop. Nobody presses record; everybody checks the light.
A piezo disc taped to the anvil closes the circuit 400 microseconds after contact. The camera keeps what it already had — we always set 30 % pre-trigger, because the interesting part begins before anything looks like it has happened.
1.9 milliseconds in, the crack front has crossed the pane and the fragments are still in formation. This is the frame that ends up as the thumbnail, and it is frame 104 of a take that runs to 9 800.
Four seconds of real time, 72 GB off the head, eleven minutes to a drive while we sweep the box. Then we do it again, because the first take is a light test whatever anybody says beforehand.
Forty seconds of screen time out of one second of hall. The witness camera lives here: it is what we send the client the same evening so they can see the rig worked.
Witness body KL-2 · full 1920 × 1080 · 2.1 TB per shoot daySeven and a half minutes out of one second. Our default. Ice, water, timber and anything that breaks in pieces you can still name are all shot here, at full sensor height.
ARC-9 at 2048 × 1536 · shutter 45° · 11.4 s of RAMThirty-seven minutes out of one second, and the first rate where light becomes the problem rather than the camera. Four 18 kW heads, every one of them flicker-free, every one of them hot enough to read by at nine metres.
ARC-9 at 896 × 512 · shutter 11° · 92 kW draw · 400 A supplyTwo hours and ten minutes out of one second, on a sensor the size of a postage stamp. We charge extra for this not because of the camera but because of the five rehearsals it takes to put the event inside a 0.9 second window.
ARC-9 at 256 × 128 · shutter 1.1 µs · 0.9 s record windowFrame rate decides how many pictures you get. Shutter angle decides how long each one stays open — and therefore how much of the event is smeared across it. Every job is an argument between these two numbers and the lighting bill.
Read it left to right: every time the wedge closes, the picture gets sharper and the hall gets darker. We shoot 11° about four times a year, and every one of those days costs the client a generator.
Nothing on this site is a simulation. Every frame came off a sensor in a cold hall in Szczecin, with two people arguing about light and a bucket of swept glass at the end of it.
Ola Rzepecka & Tomasz Byk