The tools for making images have become more accessible, more sophisticated and more capable. AI makes it possible to generate people, places and entire campaigns in a fraction of the time those things once took. Work that previously required specialist workflows is now within reach of almost anyone.

And yet, for all that new possibility, the imagery itself seems to be becoming strangely more uniform. Beautiful people moving very slowly through beautiful spaces. Golden light. Soft depth of field. Existential gazes through glass. An extraordinary number of full linen outfits. And, increasingly, the same hand touching a surface.

The technology is expanding incredibly quickly. The visual language around it seems to be doing the opposite.

That contradiction became the starting point for Rose Street. If making a polished image was becoming easier, we became more interested in what shaped it: the decisions, observations and peculiarities that made it feel authored rather than simply generated.

Rather than inherit the emerging conventions of AI-generated imagery, we wanted to make those choices in response to particular people, a particular place and a story.

Architectural imagery is usually asked to do a very specific job: explain the building clearly, show the materiality, establish the view, communicate how the spaces connect. All of that still matters. But we became interested in what might happen if not every image had to explain everything.

What if some images were allowed to create attention rather than provide information? What if a campaign could spend time on the people who might live there, the neighbourhood outside, or a detail that has almost nothing to do with the apartment itself? The feeling of a place could develop through small observations alongside the views that explain the property.

Rose Street became our attempt to explore that idea: a real Saturday in Fitzroy, gradually leading into a home that does not yet exist. The architecture would remain precise. The campaign around it could be much less obedient.

The interior was designed by Two-O Seven, and every interior view in Rose Street was rendered from our own 3D model.

We didn’t make one finished render and ask AI to extrapolate the rest of the apartment from it. It can produce remarkably convincing variations, but convincing is not the same as correct. Proportions shift, perspective becomes inconsistent and openings move. Spatial relationships become slightly more convenient. Sometimes the result looks entirely plausible until you compare it with the actual design.

Those differences might not always be obvious at first glance, but they matter to us. The 3D model remained the source of truth throughout the project. If we needed another architectural view, we composed and rendered another architectural view.

An image could be tightly cropped, partly obstructed or more interested in a person than the room around them. Whatever architecture appeared in the frame still had to be the architecture we designed.

We used AI to develop the two recurring characters, working through references, revisions and increasingly small decisions. They needed to feel specific enough to belong to this particular campaign, rather than to a general idea of contemporary inner-city living.

Their appearance was developed piece by piece: the cut of a sweater, a haircut, jewellery, rings, socks and shoes. We paid attention to hands, tattoos, proportions and the small details that needed to remain consistent from one image to the next. AI also assisted with integrating the characters into the photographs and renders, with compositing and retouching bringing everything together.

Their wardrobe was our playful interpretation of Fitzroy style, drawn from things we’d noticed and combinations we wanted to try. The black backless loafers with bright white socks were one of the details we had the most fun choosing.

The aim wasn’t character consistency for its own sake. We wanted the people to feel observed rather than assembled from familiar visual shorthand. Their behaviour mattered for the same reason. They queue for croissants, carry them home, put on a record, sit on the balcony and talk. Nothing particularly dramatic happens, and nobody needs to spend the afternoon interacting with the finishes. They simply needed to feel like the same two people continuing through the same Saturday.

The same logic applied to the neighbourhood. We didn’t want to generate an approximation of Fitzroy from everything the internet associates with “inner-city Melbourne”. We wanted the actual place, with its weather, habits and awkward bits of pavement, so we photographed it.

The neighbourhood scenes in Rose Street were built around real film photographs taken in and around Fitzroy, using both a Leica MP and a medium-format Hasselblad. We photographed the ordinary things that make the neighbourhood recognisable: bluestone, bins, graffiti, queues, shopfronts, pigeons, murals and people getting on with their Saturday.

Some of those observations became part of the story. The queue outside Lune is a little ridiculous, but also completely normal there. The characters buy croissants and carry them home. A mural they pass on the street later appears in the view from their balcony. At one point, human footsteps are echoed by pigeons walking along the kerb. There was a lot of pleasure in finding those connections, and room to have some fun with them. They gave the campaign a world outside the property, allowing the interior to feel like a continuation of somewhere real.

The photographs also brought things we could never have arrived at by simply asking for “Fitzroy”: imperfect light, accidental compositions and all the small visual information that comes from standing somewhere and noticing what is actually there. We wanted Rose Street to feel as though it could only have happened here.

Once the photographic language was established, the renders had to respond to it. We wanted the CG to sit naturally beside the photographs, not the other way around. The 2:3 portrait format of the 35 mm photographs was carried through the rendered stills and moving image, even when it made the architecture harder to compose. Following the same visual logic sometimes meant more ceiling, more foreground or less information about the room.

Light, colour and exposure were guided by the photographs rather than the conventions of a warm, glowing hero shot. The apartment needed to feel plausible within that Melbourne Saturday, with its ordinary daylight and contrast. That also meant taking some perfection back out: introducing grain, softening edges and grading the renders with the film photography in mind. Views through windows could be slightly overexposed or fall gently out of focus, as they naturally might in a photograph.

The location shaped the imagined interior too. Views, orientation and atmosphere had to make sense as a continuation of the neighbourhood outside, rather than existing as convenient backdrops. The point was never to disguise the renders as photographs, but to make the real neighbourhood and unbuilt home feel as though they belonged to the same world. A render could work beautifully on its own and still feel out of place beside the preceding street scene. In that sense, the photography stopped being reference material and became a set of rules.

Once the still images began to move, a different set of problems appeared. We chose to keep every camera static, with movement coming from the characters and their actions rather than the camera itself.

AI animation was very good at making something happen. Getting it to make the particular thing we asked for was another matter.

“Keep walking in the same direction” could end in an unexplained U-turn. A small nod could become an enthusiastic shrug. A character asked to put on a record could instead inspect it at length, rotating it in both hands as though something important had been discovered there.

So the instructions became increasingly specific: where someone was looking, how far their head could turn, which hand moved, what stayed still, and when an action started and stopped. Often we were trying to remove behaviour rather than add it, and the smallest, most natural-looking actions could take the most work.

Continuity mattered too. The characters had to remain recognisably themselves from shot to shot, with expressions and behaviour that still felt like the same two people established in the stills.

The edit determined how much of each moment the viewer needed to see and when to cut away. A technically convincing clip still had to make sense within the sequence.