A space we can discuss

The client needed to see a future office

The brief was to turn an existing space into an office. The client needed a concept for arranging the workspace, using the mezzanine and connecting it with the adjacent private office. At this stage, the proposal had to be understandable to the people who would use the space.

I used AI across several parts of the work: preparing an initial geometric model, processing the revised model, assigning materials, making visualisations, producing plans and building a browser presentation. I set the design direction, developed the layout, requested corrections and selected the results. This case follows that sequence through to what the client received.

Start by knowing what the model can prove

The initial reference was a plan image with room areas, without a complete set of dimensions. It was enough to build a provisional model with explicit assumptions. Lengths inferred from the proportions of a drawing cannot be treated as a measured survey.

AI helped prepare a SketchUp model with walls, floors and openings. It was a starting point for development. I then revised the model in SketchUp; the final plans used my measured model. These were distinct stages with different levels of confidence in their inputs.

A useful first step for a designer is to separate confirmed measurements from estimates before asking for visualisations. A convincing image does not make approximate geometry accurate.

The layout remained a design decision

The project covered the workspace with its mezzanine and the adjoining private office. The office retained the two neighbouring desks in the positions I had set. The stair, platform, furniture and circulation needed to stay consistent through every conversion.

During the work, AI mistakenly treated a distant room as the intended office and added objects there. I clarified the project boundary: those additions were removed and the camera was redirected to the adjacent room. Clear scope and a check against the source model mattered more than a persuasive description of the scene.

Another correction removed a white plane above the staircase that obstructed the opening. Geometry was checked after its removal. Errors like this belong in the model correction stage, before the final presentation is assembled.

Materials came from the actual interior

A photograph of the existing interior served as a material reference. White painted brick, grey-beige flooring, black steel, timber and a concrete ceiling established the shared palette. Residential furniture and people in the photograph were not carried into the office concept.

The model received materials and textures through Blender. The browser scene offered three lighting scenarios: daylight, general lighting and desk lighting. My revisions lowered the pendants over the workstations, introduced a ring pendant above the round table and added books to the shelving.

Lighting controls support a conversation about atmosphere and emphasis. I do not present them as a lighting calculation or a completed fixture specification. Those require their own inputs and checks.

Seven views built from the model

Photorealistic images were created with imagegen using selected views of the 3D model. The model view supplied the camera, geometry and furniture positions; the reference photograph and previously selected images helped maintain material and visual continuity.

The final set contained seven views: the workspace, the mezzanine with books, the staircase, a reverse view from the platform, the adjacent office and two views beneath the mezzanine. One lower view was widened so a desk could fit completely inside the frame. Selection involved checking what the client could actually see, as well as image quality.

These are generative concept visualisations. Small details and proportions may differ from the model. Dimensions should come from geometry and plans; images support discussion of atmosphere, surfaces and composition.

Plans added another level of definition

After the visualisations, the project needed dimensioned plans. Four A3 sheets were prepared from the revised model: first-level and mezzanine survey plans, followed by furniture layouts for both levels. The survey sheets used scales of 1:50 and 1:25.

Furniture sheets added item references, overall dimensions and offsets to the internal wall faces. Where chair outlines obscured the desktops, the symbols were refined and portions beneath the desks were hidden. Plants were labelled as indoor plants in pots.

PDF supports viewing and printing; DXF supports further editing. The dimensions were extracted from my measured model. This set describes the concept shown here. A full construction package, mezzanine structure and building services still need their own development.

A space the client could explore

The deliverable was assembled as a browser presentation. It lets the viewer rotate the model, select views and lighting, adjust the cut height, show the ceiling and proposed details, and save images. Visualisations, plans and PDF/DXF downloads sit below the model.

The interface needed revision too. At first, the mouse wheel zoomed the model and made it difficult to reach the gallery. The model was placed in its own frame, normal page scrolling was restored and zooming moved to buttons. The opening cut height was set to 4.25 metres to show the full space immediately.

The format gives the client several ways into the same proposal: a visualisation for the overall impression, an interactive model for a chosen viewpoint, and a plan for dimensions. Viewing the prepared presentation does not require opening the source model in SketchUp.

A proposal the client decided to use

The client needed a concept for fitting out a future office. I delivered the interactive presentation by link. The feedback I received described a wow effect; the proposal was taken forward as the basis for the planned office fit-out.

At the time of writing, this is an accepted concept rather than a completed interior. There is no evidence here for a specific budget saving, a quantified reduction in delivery time or an operational result. The supported outcome is a clear proposal that the client chose to develop further.

Where AI assistance becomes useful

In this project, AI helped translate the same design intention between a diagram, a model, materials, images, drawings and an interactive page. Each step became useful when its output could be checked and carried into the next one.

For a designer, that means defining the project boundary, naming the authoritative geometry, choosing a relevant view, checking changes and retaining editable files. For the client, it means a proposal they can inspect and discuss before fitting out the space.

For a similar workflow review or interactive presentation of your project, we can start with the available materials and what your client needs to understand. See the working formats, the Method or website development for a dedicated digital presentation.