THERE’S A SENSE OF SATISFACTION when you finish the perfect packing job, whether it’s a container or a suitcase. Kapitza know that moment when you step back in admiration after carefully squeezing that final piece into a tight space – or jam it in.
Jam – as musicians from Dave Grohl to Miles Davis would attest – is also the free-flowing collaboration where artists improvise and let something completely new find form.
The two meanings intersected when Decisive Agents challenged the artist duo Kapitza to design a set of their trademark day-glo boxes, and then work with AI platform DeepPack to squeeze them into a standard shipping container. The idea was to collaborate with the software, tuning the dimensions of the boxes to provoke new and challenging configurations. The resulting artwork is Jam. And it asks new questions about AI and its creative process.
Who are Kapitza?
Sisters Petra and Nicole Kapitza grew up in rural Germany and are now based in East London. They say they’re on a mission to colour the world with positive and uplifting works of art, where colours collide with mathematics to produce vivid and minimalist geometric designs.
Kapitza has exhibited in Europe, Japan and the United States. They create bold, large-scale murals, uplifting installations, dynamic public art, intelligent textiles, colourful products and also publish books and run a print studio. They’ve collaborated with international brands Ikea, Swatch and Clinique and institutions like the University of Cambridge, Royal London Hospital and the U.S. Postal Service.
Against the backdrop of an ongoing debate around AI’s potential impact on the creative industry, Kapitza welcomed the chance to play with such a unique tool. Deep-Pack is AI-powered 3D packing software designed to create optimal load plans for transported goods. It essentially tells you how to pack your stuff to ensure as little wasted space as possible, while considering things like stackability and what you need to unload first.
Unlike generative AI such as MidJourney or Dalle AI, DeepPack is a logistics tool that uses reinforcement learning to optimise cargo planning. We wanted to know what would happen when the AI was pushed to think outside the box.
We asked Kapitza to pack a shipment of their iconic day-glo boxes, playing with size, colour and load requirements to trigger new and original combinations. They worked with the most common shipping container size of 20ft x 40ft, but there are lots of other containers, air and ground pallets and ULDs preprogrammed into the software.
Creating dimensions of boxes
The first step for Kapitza was to define the box dimensions. As a shipper, you wouldn’t need to do this because the boxes would already be a certain size. But creatively, they were coming at it from a different angle. They downloaded the packing sheet and played around with measurements, initially trying to get the container filled up neatly. “I made my brief to myself: How can I fill the container so there’s as little space as possible left? Then it was about me doing it mathematically, working out the sizes I needed to fill it up.”
Feeding dimensions into DeepPack
The dimensions from the packing sheet were then fed into the software. This is where the colour comes into it. “One of my packing sheets was really, really, really long and it just came up with another colour and another colour…I like that randomness of the colour palette.” Essentially, DeepPack uses colour to group and differentiate the boxes. And these colours are key because it’s this that brings DeepPack’s 3D output to life, increasing the usability of the final visualisation.
Once Kapitza had input and run the software, they had the first 3D visualisation of their loaded container. DeepPack will always produce the most efficient solution possible for that load and container combination, and adjust for the specific needs of the shipment. The initial visual reflected this palpable logic. Now Kapitza began to change the box measurements to watch how the AI would respond. “I started coming at it from more of a numbers side. For me, maths is always interesting.”
Kapitza tries to make AI improvise
DeepPack kept putting the same-size boxes next to each other, insisting on some sense of order that was at odds with Kapitza’s vision. “I wanted them to be more random. So I was trying to make it less efficient!” Adjusting the parameters provoked new iterations, but they noticed that it would always pack bigger things on the outside, meaning smaller boxes would disappear from the artwork. “Effectively, you’re almost like changing the dimensions of boxes to have nice rhythm.”
One of the biggest challenges was filling up the space entirely, which Kapitza wanted to trigger this ‘jam’ quality in the visual. But with DeepPack, if a box doesn’t logically fit into that container, it’ll add it to the next one.
Weighing up realism with creative output
Kapitza wanted their boxes to feel and behave like real cargo. But to both fill the container and trigger random combinations, some of the boxes got very small or very slim.“I thought, is this realistic? But then I thought… it could be tops of IKEA tables.”
In a real shipping scenario, these smaller boxes, like table tops or kitchenware, would likely be put inside a larger box. “I was trying to be realistic but then also use my creative license to making smaller ones. I thought if that looks better in the composition, that’s great.”
Applying the final Kapitza touch
By creating boxes with dimensions that achieved rhythm and contrast, without being split into separate containers, Kapitza got a composition they could work with. They then laid it out using Adobe Dimension, as though looking at the container side-on. The final stage was to apply Kapitza colours, specific shades curated for effect, and use shadowing to mimic the Tetris-like tangibility of DeepPack’s output.
“The work we do is often very mathematical. Normally, as a box might get larger in size within a square, I know aesthetically where I want to put it. Whereas when the AI is placing the boxes, I had no control where it’s going.”
The collaborative process with DeepPack had many similarities to Kapitza’s signature approach, which usually starts with a simple geometric shape that will be rotated, split or multiplied as a particular logic is applied to it. And as with any of Kapitza’s artworks, there is both tension and flow. While the software insisted on asserting its logic, Kapitza responded with their mathematical approach and empathy for DeepPack’s drive for efficiency. “I removed some boxes in the middle, which looked really interesting. But I thought… you’re not gonna be happy with that because that means it’s not efficient, but it looked nice.”
