Skip to content

Projects with no playbook

An exhibition whose pieces are scattered across the world. Trains nobody is allowed near. An avenue that will not be built for another two years. A closed motor show. Four times the same situation: nothing in the catalogue answered. Each case also says what the artificial-intelligence layer changes in the making of it.

Tell us about your project

Four cases, and what AI changes

Bozar

Cultural institution

A virtual exhibition room: a complete lamellar armour at the centre of a plinth, a helmet and boots around it, on a herringbone parquet floor.
A rebuilt room, with its objects put back in place and its information markers.

The problem

Mounting an exhibition whose pieces are scattered across the world, and part of whose archive has disappeared. The objects could neither travel nor, in some cases, even be photographed from every angle: the existing catalogues are two-dimensional, and nobody had seen the back of certain pieces.

The answer

The digitised catalogues of the previous exhibition were analysed to identify what was missing, then the pieces were rebuilt in three dimensions and through 360 degrees, under the supervision of archaeologists. The museum's rooms were surveyed by laser, which gave the exact setting in which to place those objects, and the exhibition « A Passage to Asia » became visitable from a browser.

What followed

The result was convincing enough that visitors turned up at the museum's ticket desk and asked how to get into the exhibition.

The AI layer

Working through the catalogues was the expensive part: hundreds of pages to cross-check in order to know what is documented and what is not. A model now reads those pages, picks out the dimensions, the materials and the missing faces, and flags the sources that contradict one another. What it proposes to fill a gap is never used as it stands: the decision stays with the archaeologist, and that is what lets a museum sign off on what we deliver.

Another virtual room, with a bronze bell tipped on its mount, a stone jar and two Gandhara statues.
The laser survey carries the proportions and the light of the real room.

Thalys (now Eurostar)

Rail transport

A red Thalys train at a station platform, under overhead lines, with the number 4341 on its nose.
The train at the platform, rebuilt entirely in computer graphics.

The problem

Presenting train interiors designed by Matali Crasset to the international press, when access to the production sites was limited and no train was available for a physical reveal. An object that could not be shown had to be unveiled.

The answer

The trains were rebuilt in a digital station, to the level of detail the specialist press was going to inspect: the position of every screw, the exact colour shades, the seat dimensions to the centimetre. The platform ran in a browser, on desktop, on a phone and in virtual reality.

What followed

The press was able to walk through the trains live on the day of the announcement, and to come back to them afterwards.

The AI layer

The same train had to run in a browser, on a phone and in virtual reality, where dropping below 90 frames a second makes people queasy. Several levels of detail therefore had to come out of a single manufacturer's model. Texture upscaling and render denoising are handed to learning-based tools: each iteration falls from hours to minutes, and that is what makes three targets affordable inside the timetable of a press announcement.

Inside the on-board kiosk: a lacquered red counter, round shelves, a bench seat and a band of windows.
The on-board kiosk, one of the spaces you can walk into from the station.

STIB / MIVB

Public transport

A view of the redeveloped avenue: paving, wooden benches, flower beds, bicycle stands, passers-by, and a bus in the distance. A navigation bar offers four zones.
The avenue as it will be, walkable zone by zone.

The problem

Winning residents over to the redevelopment of a Brussels avenue. The project existed only as technical drawings, unreadable to anyone outside the trade, and roadworks are usually judged on the disruption they cause rather than on what they will bring.

The answer

The drawings were turned into a walk at eye level, zone by zone, moving through the avenue as it will be: the benches, the trees, the cycle lanes, the stops. Each point of interest explains what changes at that precise spot. The walk was distributed to residents, in French and in Dutch.

What followed

The tool became the basis for discussions between the project's stakeholders, who could compare their intentions on a single image rather than on drawings.

The AI layer

The projected avenue had to sit inside the real one: the façades opposite, the trees already there, the light of the place. That reconstruction comes from photographic surveys processed by photogrammetry, where matching the views and cleaning the point clouds are learning-based steps. Planting, street furniture and passers-by are generated and then corrected rather than placed one by one. Without that, a whole avenue, in four zones and two languages, does not fit the budget of a public consultation.

D’Ieteren

Car distribution

A digital motor show: a blue CUPRA Formentor on a turning plinth against a sunset, two other models waiting, and buttons for live advice, configurator and terms.
Each model on its plinth, examinable through 360 degrees, with the adviser one click away.

The problem

Selling cars while the Brussels Motor Show was closed, at the point in the year when a decisive share of orders is settled. Belgium's leading car importer found itself without the place where its customers come to look, compare and decide.

The answer

A digital show was built, where each model can be examined in three dimensions, outside and inside, with its specification and its terms. Visitors found a live adviser there, a configurator, and the option of booking a video appointment or a test drive.

What followed

The setup made it possible to keep receiving qualified enquiries while the show was closed, and to reach buyers who would not have travelled.

The AI layer

A digital show is first of all a catalogue: every model, every trim, every colour, seen from several angles inside and out. The count runs into thousands of images, and the show was closed now, not in six months. Learning-based denoising divides the computing time for each view, which makes it possible to produce the full range rather than a selection. Human eyes then go back over what the machine gets wrong most: reflections on dark bodywork.

What it changes for an ordinary project

None of these four projects can be ordered from a catalogue, and that is not what we are offering here. What they have in common is one thing that serves every day: when the subject cannot be photographed, it has to be made.

It is the same work as showing an office floor that is still empty, a product that does not hold up in a photograph, or a place the customer cannot travel to. The budget changes, the question does not. Cases like those are handled in a website build or a rebuild, like everything else.

Something you cannot manage to show?

Tell us about your project