Not Quite Past offers culturally-informed, crafted AI that is linked up with high-quality suppliers. All we need to do is generate an image, from any prompt, in the Workshop or select from pre-existing designs in the Marketplace, and our tiles get delivered to our door. The result is a revolutionary approach to design. Rapidly create highly personalised, one-of-a-kind artwork that plays with the rich cultural traditions of the past. 'Tiles made by you with AI.' We spoke to the founders Adam Davies and Jack Marsh about their inspiration, the relationship between AI and creativity and crafting AI in the visual heritage of the Dutch Golden Age.
1. Where did the idea behind Not Quite Past come from?
Jack: The short story is this: Adam and I do these architectural visits (with these books from the Buildings of England series called Pevsners). We were on one of our architectural visits around the country, this time to Birmingham.
In the Victorian era, it was a powerhouse of design and manufacturing. This era is seen everywhere in glorious factories and grand civic buildings. Birmingham was a centre of high-quality manufacturing—amongst the most advanced in the world at its prime–specialising in guns, textiles, toys, brass and other things—typically goods made of many small, fine components.
Following the Industrial Revolution, many people were dissatisfied with the cheap quality and crude design of manufactured goods. There was a huge push to raise the standard and aesthetics of manufactured goods, to blend master-craftsmanship with mass production.
This was, for instance, part of the motivation for the creation of the Victoria and Albert Museum in South Kensington. When it was founded in 1852 by Henry Cole, it was actually called the Museum of Manufactures. The V&A was meant to be an encyclopaedic repository of all the great crafts and artistic traditions of the world. They wanted artisans and captains of industry to come to the museum, be inspired by the grandeur and beauty of past works, study them intensively, and then spread what they saw and learnt into the productions of the day. That’s why there are those two extraordinary “Cast Courts” in the V&A: having the original artefact was not entirely necessary (contrary to what we might assume about museums today), since, for their purposes, all they needed were the exact forms from which to study.
Let me get back to Birmingham!
So we had come to see the famous red brick neo-Gothic and Arts and Crafts buildings of Birmingham, but it was outside the Bell Edison building that our idea first came to us. There is this incredible terracotta brick gable ornament to the top. It was complex, beautiful and also mass produced. I thought, what happened to mass-produced ornament on our buildings? Surely something like this could easily be made in our day, but even easier? But ornament is barely present in contemporary building. We have much more advanced techniques since the 1800s, and yet our ornament has not kept pace at all. It feels as if ornament exists in a state of banishment, while a tyranny of glass, metal, white unmoulded walls, or brick New London Vernacular towers carries on virtually unchallenged.
This discussion about industrial production, ornament, and the modernist built environment continued, amplified by each Victorian red-brick building we saw, covered in highly complex, rooted, playful ornament.
Adam: I have a background in tech, and Jack did art history at university and had worked as an intern at an AI art fair a few years before this. While we were walking around, we started wondering whether we might be at the cusp of a similar (but arguably even bigger) moment in the history of technology to the Victorian era. Maybe AI could reduce the cost of designing architectural ornament just as these Victorian buildings benefited from the falling cost of producing architectural ornament with the Industrial Revolution?
Jack: While I was playing around with different kinds of architectural ornament on an AI image generator, for some reason I still don’t understand, I decided to ask it to produce a Delftware tile. William Morris, for example, had been interested in these spare Dutch tiles from their golden age, so there’s quite a bit of heritage there. The result was fantastic, though it needed some training. But it immediately seized me as an idea.
It was particularly beautiful because the output of the AI image generator is itself square, and of course tiles are square. So, if we connected it up with a supplier, it would almost be like that square on the computer screen hardening up and falling into your hands!
Delftware tiles work well both as regards both subject and style.
Subject: Delftware tiles were filled with crazy subject matter. The Dutch literally put everything and anything they saw or dreamed up into those tiles: biblical scenes to defecating cupids. It seemed to share the same exploratory, playful spirit that AI prompting invites.
Style: many delftware tiles, by the mid-1600s onwards, had a beautiful and spare style, lots of white background surrounding and isolating a single figure, with four reduced corner motifs. This structure works brilliantly with the current state of AI. There is a strong figure-ground relationship that lends itself to prompting with words.
But, as with any idea, there’s the short story of the specific trigger and the long story of slow, subconscious combinations working over months and years. Adam and I have definitely been circling around this idea for years. It feels geyser-like.
2. What excites you about working with AI ?
Adam: It sounds a little trite, but I think it is true: technological change is pushing us into fundamentally different conditions, and we have a choice: either to resist or embrace. For the first time in human history, it will be not just possible but extremely easy for people to create beautiful things that they design themselves.
But what we want to do is actually a little different from what people might expect. We actually want to root this technological change in the traditions of the past. AI is trained on the past. That’s what it knows. And we found that it can produce past objects with quite compelling fidelity. We want AI to support and explore artistic traditions, not blast them apart.
Jack: As in Birmingham, comparing the Victorian ornament with the blank modern, it seemed to us that industrial modernity has slowly brought about homogeneity and boredom and, certainly in the past, a cultural imperialism of a sort: buildings now seem to look the same everywhere.
We really think the AI revolution has the distinct capacity to reintroduce and support local traditions, identities, histories back into design and architecture. We’re incredibly excited about this, about the prospect of helping root state-of-the-art innovation in an art historical context. We want to override decision by committee, international sameness, the lowest common denominator of profit-motives, and instead to encourage a new kind of artificial intelligence Arts and Crafts movement.
3. How does Not Quite Past balance AI’s automation of the creative process with the human/humanity’s creative skills? And how important do you think this human involvement is?
Jack: People think you should demand what you want from the prompt, and then get it. But that isn’t what always happens – you don’t always get what you want. And I think that is a great thing. I have been amazed at how much of my prompting in the AI is discovery, rather than a simple transaction. The experience is never: I have this exact idea in my head, and this is what I want to get. But, it is more: wow, this is a really wacky prompt, and I can’t wait to see what I get.
Figures 3, 4, 5, 6: Example of wacky results or wacky prompts. 1. Prompt: A hot air balloon over the Bristol Channel. No Bristol Channel, but a beautiful image instead. – 2. Prompt: A child flying a colourful kite on a breezy beach. – The child is utterly taken over by the wind and the kite, so it seems. 3. Prompt: delftware tile of a dark, lush, red and gold interior of opera house, red velvet drapery, stage with singers, high up in one of the prestigious opera boxes is sat a glowing yellow rubber duck. 4. Prompt: Marc Chagall minimalist painting with bright colours of night-time landscape with stars and flying humans.
So much of the creative power of the AI generator, for me, at least, has been the strange stylistic combinations I have forced through, generating quite unusual, unprecedented looks. I have created a whole series of Delftware Kandinsky hybrids, for instance, which are stunningly beautiful.
I have a line: the prompt is only the digger’s chisel. We are almost on an archaeological dig when we are searching through our model. When prompting our model, we are on this vast pile of archaeological treasures. The prompt we use is actually just a tool to dig some beautiful artefact up from the ground, previously hidden. We never really know what we find until we dig.
More importantly, we are the ones who decide what is beautiful, what has meaning. Not the AI. Human involvement is ultimately the only thing that makes AI imagery work. We find meaning.
Adam: Just to add, we’re also not just using an off-the-shelf AI model that makes just anything. We’ve crafted it to be rooted in the particular visual heritage of the Dutch Golden Age, honouring the contributions of generations of human craftsmen who pioneered that beautiful blue and white style.
Jack: The patron and the artisan are blurring into each other, eliding into one another.
4. Top three favourite prompts so far?
Jack: This is a tough question because Adam and I have been looking at these for some months, so we have a real taste for the exotic. When we first started, the more normal ones were astounding. But maybe we have become a bit jaded and need harder stuff! Anyway:
“80s arcade game glowing neon in the darkness pixelated game creatures, spare”
That is bonkers, and I love it. But, you see, it’s blending delftware-ness with 80s arcade game aesthetics and the word ‘neon’ and all that it connotes. But look in the corners, and you will see the traditional delftware “hoekmotieven” or corner motifs. It’s still, residually, a delftware tile in structure to some degree. That’s the sort of wild combinations you can achieve.
Adam: That one is really mad. Here are two more typical, but still very beautiful ones:
“moose playing hockey” — Animals are always really fun to depict, and the AI model tends to render them quite well, as long as you carefully count the number of legs! This is quite a compelling design and very wacky.
“a space shuttle” — We recently launched a new feature that allows for specific AI models to be used for different corner motifs. One of my favourite ones is a fleur-de-lys model that puts that symbol in the corner and gives the whole tile a brushier, rougher look. This pairs super well with the design of a space shuttle in the centre of the image, giving it a fun and quirky look.
5. What has surprised you most about working on this project & how is that informing how you are developing the project/future projects?
Adam: Neither of us comes from a manufacturing background — I worked in software while Jack has spent some of his career in the art world, and some in education.
One of the biggest things we’ve learned is how important the nitty-gritty of materials can be. AI models can generate any colour on a screen, even ones that cannot actually be produced with ceramic pigments or survive the firing process.
I visited our tile factory supplier, and speaking to the skilled craftsmen who make our tiles was an eye-opening experience. We learned that we had to carefully design our software to work well with the characteristics of our materials to produce truly beautiful and durable products.
Jack: I completely agree. Material is crucial.
One thing else I really am keen to work on is explore the medium of tiles. I think there is so much potential in the form and shape and purpose. One of the projects I am working on is exploring how tiles do not have to be fixed onto the wall, static, but moving, floating free, exchangeable. Hopefully it will come to fruition soon!