Most people who try Strukt AI for the first time don't fully know what to expect. They upload a floor plan, something happens, and a 3D model appears. What's less obvious is what's actually happening at each stage — and why each step is distinct from the one before it. This is the full walkthrough.
Step 1: Upload your floor plan
The process starts with any 2D floor plan. That can be a PDF from an architect, a DWG export from AutoCAD, a PNG scan of a printed drawing, or even a clear photograph of a hand-sketched plan. There's no prescribed format.
When you upload, the first thing the AI does is read the geometry — it finds the wall lines, detects wall thickness, identifies openings (doors and windows), and locates room boundaries. Broken or incomplete lines get repaired. Dimensions are extracted where readable, and a calibration step lets you anchor one known measurement (like a door width) to fix the real-world scale of the entire model.
The result of Step 1 is a digitized 2D model — a structured representation of the floor plan with detected wall positions, room polygons, and opening locations. This is the foundation everything else is built on.
Step 2: AI generates the 3D structure
The 2D model gets extruded into 3D. This is the step that feels like magic but is actually a precise spatial operation: every wall polygon becomes a 3D volume at the correct thickness; every opening becomes a void cut through the wall geometry; every room becomes a navigable enclosed space with a floor and ceiling.
The output is a live 3D model — not a rendered image of a 3D model, but the geometry itself. You can see it from any angle in the editor. You can click on walls and move them. You can adjust opening positions. You can add or remove furniture.
This is the part that separates spatial AI from image-generation AI. Midjourney can produce a beautiful picture of a room. What it can't do is give you a wall you can move 30cm to the left and immediately see the effect on every adjacent room. The 3D structure that comes out of Step 2 can do that.
At this stage you're also working in 2D top-down view — the flat plan perspective that designers are used to — which makes it easy to adjust layouts, snap doors to walls, and place furniture by dragging.
Step 3: Walk through in first-person view
Once the structure looks right, you switch to first-person view (FPV). The camera drops to eye level inside the model and you navigate through the space like a first-person video game — walking through rooms, looking through doorways, standing at the kitchen island to see the view across the living room.
This step is less about editing and more about feeling the space. Proportions that look fine on a 2D plan can feel wrong at eye level — a corridor that seemed adequate at 900mm turns out to feel cramped when you walk through it, a living room that seemed generous turns out to have an awkward relationship with the window position.
FPV is also where clients become genuinely engaged. A 2D plan is a professional document — clients can study it but struggle to inhabit it. A first-person walkthrough of their future space, furnished and lit, is immediately legible to anyone regardless of design experience. Decisions that would take three revision rounds on a 2D plan get made in a single walkthrough.
Step 4: Apply materials and render
The final step is the one that produces the shareable output. You select materials from real brand catalogs — actual tile products, actual laminate ranges, actual paint colours — and apply them to surfaces. The materials use PBR (Physically Based Rendering) maps: normal maps for surface texture, roughness maps for how light scatters, metalness maps where relevant. What you see in the editor is physically accurate.
Lighting is set — natural light direction, time of day, ambient intensity — and then you render. The render engine takes the scene geometry, the material properties, and the lighting parameters and produces a photorealistic image. This is what you share with a client, include in a proposal, or post as a project preview.
The key thing about the render: because the geometry and materials are live, you can render the same space a dozen times with different configurations. Morning light vs. evening light. Timber floor vs. polished concrete. Dark kitchen cabinets vs. white. Each render takes seconds and there's no rebuilding the scene between them.
The full pipeline in one view
| Step | What happens | Output |
|---|---|---|
| 01 — Upload | AI reads walls, openings, rooms. Dimensions calibrated. | Digitized 2D model |
| 02 — 3D Build | Walls extrude to volumes. Doors/windows cut openings. Rooms enclosed. | Live editable 3D model |
| 03 — FPV | Camera at eye level. Navigate rooms. Check proportions and flow. | Spatial validation |
| 04 — Render | Real brand materials applied. Lighting set. Photorealistic image generated. | Shareable render |
How long does it actually take?
For a standard single-floor residential plan: Step 1 (upload and detection) takes about 15–30 seconds. Step 2 (3D generation) takes 30–90 seconds depending on complexity. Steps 3 and 4 are user-paced — you spend as long in FPV and material selection as you want. The final render takes 5–15 seconds.
From upload to first render: typically under two minutes for a clean residential plan. The bottleneck is usually the designer making decisions about materials, not the processing time.
What makes a good input plan?
The quality of the 3D model correlates directly with the quality of the input. The things that help most:
- →Clean, unambiguous wall lines — no overlapping or broken segments
- →Doors and windows clearly indicated with standard symbols
- →Cropped tightly — no title blocks, borders, or annotation clouds in the upload
- →At least one known dimension for accurate calibration
- →Good contrast — dark lines on a light background
A CAD-derived PDF will generally produce better results than a smartphone photo of a printed plan, but the system handles both. If the first pass misses something, the 2D editor lets you add, move, or delete walls manually to correct the model before generating the 3D.
Try it
The best way to understand the pipeline is to run it. Go to app.struktai.work, upload any floor plan you have to hand — even a rough sketch — and follow it through all four steps. The free tier includes the full pipeline with no time limit.