Vera van der Burg puts AI and ceramics into feedback loop

Van der burg — a quick note to anchor this piece for readers.

Van der burg: Quick notes

Dutch designer Vera van der Burg aimed to “slow down” AI by combining it with a ceramics practice in the project From Text-to-Clay, which has seen her exhibit several unworldly creations.

From Text-to-Clay is part of van der Burg’s ongoing PhD research into reflective AI practices, which involved her undertaking a back-and-forth process with an AI image generator, training it on photos of her work in order to create new images.

She then replicated these in clay, only to feed them back into the model and start the process again.

Vera van der Berg worked with AI to generate images of ceramics (above) before hand building them in real life (top image)

Van der Burg was drawn to combine AI with ceramics because of their apparent difference – one seen as slow and tactile, the other instant and insubstantial.

“From Text-to-Clay is part of a larger piece of research that I’ve been doing for five or six years where I’m searching for different ways of using AI in your creative process as a designer or artist – especially as a way to not speed stuff up or automate or make it more efficient but as a way to reflect,” she said.

She also saw in ceramics an opportunity to work with the unpredictable side of AI and to treat it more as a material than a product.

The resulting sculptures mirror the renders as closely as possible

“For centuries, working with clay has taught practitioners a lesson in humility: you do not master it, you collaborate with its agency,” said van der Burg. “We approach AI, however, with an opposing desire for control and predictability.”

“I felt like the material lens could be used more in how we see AI,” the designer told Dezeen. “If you use AI as this end-based thing, it’s kind of limited, but if you see it as this assemblage of different steps, the way a ceramics practice also is – where you first have to do this, and then you have to do that for the material to develop – it’s more interesting.”

The training process is one such step that van der Burg observed was often left out of the thinking around AI in design.

Van der Berg trained and prompted using abstract terms like “ceramic female pain”

Working with the open-source image generator Stable Diffusion, she fine-tuned the model with a set of her own training data consisting of photographs of ceramic objects she had made during her student days.

Knowing that text-to-image generators rely on making connections between the language used to annotate images they are trained on with the language used later during prompting, she focused on making annotations that were personal, emotive and abstract rather than purely descriptive.

This allowed her to use prompts like “ceramic divorce”, “ceramic sadness” and “ceramic female pain” and generate images that she found to be compelling examples of each.

Eight actually interesting uses of AI from 2025

She also took care with selecting her “material”, choosing an early version of Stable Diffusion that she felt had “a very specific aesthetic”, rather than the more faultless current version.

When it came time to hand-build her generated images in clay, van der Burg found herself drawn most strongly to the ones that “felt impossible”.

Because AI image generators mimic patterns from training data but do not understand materials or physics, their inventions can be detached from the laws of reality.

The sculptures have unworldly forms

“These images had no understanding of clay’s actual making process,” said van der Burg. “Once you zoom in, there are so many mistakes.”

“You cannot really see how things are connected or if something is a dent or a shadow or a ball of something,” she continued. “Light sources, nine times out of ten, don’t make sense.”

The uncertainty and absurdity were what appealed to van der Burg, who undertook the project during a three-month residency at the European Ceramic Work Centre. “I felt that it was like a negotiation,” she said.

One of the works is based on an image of “clay ignorance”

The resulting sculptures then became more training data in van der Burg’s circular process. Using the same annotations and prompts, she generated more images and made more clay works, revelling in the increasingly distorted shapes.

“The funny thing is, it started to always invite me to go bigger or to do more,” said van der Berg. “So where the shape first had maybe four holes in it, it would give me eight holes.”

“It’s kind of like how ChatGPT always asks you, do you want more?”

The AI generated images were created through Stable Diffusion

Van der Burg exhibited her surviving sculptures – the ones that didn’t explode in the kiln or collapse under their own weight – late last year at Dutch Design Week, where she was a winner of the Emerging Talent Award.

Her emphasis on language and training was echoed by another recent AI collaboration project from Ross Lovegrove’s studio, this one involving chair design.

In that case, the team focused on specialised vocabulary as a way to emulate the designer’s signature style.