Polycam Capture Playbook — the whole-property sprint
Polycam's value is capture-time only. Every scan exported as GLB / OBJ / PLY / DXF / CSV is yours forever, and the studio's viewers are local three.js — no subscription needed after capture day. So before the Business trial ends, capture everything you might ever want a digital twin of. Worked example: a two-storey family home with a large front and back garden.
Whole house in one scan, or room by room?
The most important question, answered first — because it determines the entire walk order.
You cannot cleanly split a whole-house scan into rooms later. Polycam gives you a crop box, not a room-splitter — walls are shared surfaces, doorways smear between spaces, and long continuous walks accumulate drift (5–15 cm over a full floor loop is normal). Cutting a per-room twin out of a house mesh is painful and the result is worse than a dedicated capture.
The reverse is also true: separate room meshes cannot be reliably stitched into one house model afterwards — there is no shared coordinate frame to align them.
So each output gets its own dedicated pass. Three passes, three products — none substitutes for another.
| Pass | Mode | Doors | What it produces | Why it can't be derived from the others |
|---|---|---|---|---|
| A · Per-room mesh | LiDAR | CLOSED | Measurement-grade 3D twin of each room — the restyle asset, the viewer asset | Whole-house walks drift; cropping rooms out is messy |
| B · Whole-floor walk | LiDAR | OPEN | One connected model per floor — the "explorer" view, hallway relationships, flow | Separate room meshes can't be stitched together |
| C · Floor plans | ROOM | OPEN | Parametric plan per room (walls / doors / windows), then merged into a whole-floor plan in-app | Mesh passes give geometry, not clean architectural plans |
Every mode, with settings
Five capture modes, each with a job. Settings below are the recommended starting points for this property — adjust only if the live preview looks wrong.
LiDAR mode
Real-time depth mesh from the iPhone Pro sensor. Fast, reliable on blank walls where photogrammetry fails, accurate to roughly 1–2 cm. The workhorse for passes A and B.
Photo mode (photogrammetry)
Shoots overlapping photos, cloud-processes into a high-detail mesh. Far better than LiDAR beyond 5 m and for texture quality. Needs even light and stillness — wind-blown planting wrecks it.
Room mode (floor plans)
Apple RoomPlan: recognises walls, doors, windows and fixtures as parametric objects rather than raw mesh. One capture per room, then merge rooms into a whole-floor plan in-app.
Gaussian splats
Processed from a photo-mode-style capture into a radiance field: photoreal view synthesis meshes can't touch — reflections, soft materials, depth of field. The showpiece format for the studio's capability pack.
360 mode
Ten seconds per room from the centre point. Not geometry — but a dated, complete visual record of every surface, and the fastest insurance-grade evidence you can capture. Do it in the same visit as the mesh pass.
Room technique — the walk
Same drill for every room, so the whole set comes out consistent. Budget 6–8 minutes per room across all three captures.
Before you press record
- Lights on, all of them — LiDAR doesn't need light but the mesh texture does. Kill direct sun stripes with blinds; flat, even light wins.
- Door closed for the per-room mesh pass (stops corridor bleed) — open for Room mode and the whole-floor walk.
- Mirrors and glass — LiDAR sees through or ghosts them. Don't dwell on them; a quick pass is fine, staring at them creates phantom geometry.
- Declutter to taste — whatever is visible is in the twin forever. Thirty seconds moving laundry is cheaper than living with it in 3D.
- Storage free — meshes and photo sets are heavy; check the phone has 20 GB+ free before day one.
The sweep pattern (LiDAR mesh pass)
- Start in the corner by the door, phone at chest height, tilted slightly down.
- Walk the perimeter slowly, keeping 1–2.5 m from the surfaces you're scanning, sweeping each wall floor-to-ceiling in smooth vertical passes.
- Extra dwell in corners — they anchor the geometry.
- Second lap for ceiling and floor: tilt up, lap; tilt down, lap.
- Finish where you started — closing the loop lets the tracker reconcile drift.
- Watch the live mesh like a hawk: grey holes mean go back now, not in processing.
Stairs and the whole-floor walk
- Whole-floor pass (pass B): every door open, one continuous unhurried loop — hall first, then in-and-out of each room without stopping the capture, back to the start point.
- Stairs: a dedicated continuous capture from the bottom hallway up to the top landing, sweeping both stair walls and the treads. This is the geometric link between your two floors.
- Expect drift on long walks — that's why pass A exists. Don't chase perfection here; connectivity is the product, not precision.
The four-day plan
Two floors, two gardens, one deadline. Export the same evening as each capture day — never let the cloud hold the only copy overnight if you can help it.
Every ground-floor room, three captures each
Per room: Room mode (door open) → LiDAR mesh (door closed) → 360 pano from the centre. Include the hall, under-stairs, utility, WC — the unglamorous spaces are the ones you'll want for storage and extension planning.
Finish with the whole-ground-floor LiDAR walk (all doors open), then merge the Room-mode captures into the ground-floor plan in-app. Evening: export everything, file it.
Upstairs, same drill, plus the stair link
All bedrooms, bathroom, landing, hot press, attic hatch if accessible. Same three captures per room. Then: the stairs capture (bottom hall to top landing), the whole-first-floor walk, and the first-floor plan merge. If Polycam will merge both floors into one building plan, do that too. Evening: export, file.
Gardens, exterior, splats, furniture
Morning (still air, soft light): back garden photo-mode capture, then front garden. LiDAR pass on hard landscaping — patio, steps, walls. Exterior orbit of the house. Afternoon indoors: Gaussian splat captures of the two hero rooms, photo-mode object captures of 3–5 hero furniture pieces. Evening: export, file.
Verify, re-shoot gaps, pull everything down
Walk the export checklist against the filing tree — every capture, every format. Tape-verify one wall per floor and one garden edge. Re-scan anything with holes. Confirm the merged floor plans exported. Then section 10.
Gardens — yes, with the right mode
Big front garden, big back garden: these are photo-mode jobs with a LiDAR assist, and the conditions matter more than the technique.
Conditions (non-negotiable)
- Overcast beats sunny — even light, no hard shadows baked into the texture, no LiDAR interference from direct sun.
- Still air — moving leaves and swaying stems turn to mush in photogrammetry. Early morning is usually calmest.
- Dry surfaces — wet, reflective paving scans badly.
The capture pattern (per garden)
- Photo mode, auto-shutter. Walk the full perimeter three times: once at chest height aimed level, once aimed down ~45°, once crouched low. That's the orbit set.
- Then a grid pass over the lawn — walk stripes across the open ground, camera down ~45°, mowing-pattern overlap.
- 150–250 photos per garden. More is fine on Business processing.
- Include the house facade edges in garden shots — it anchors the garden to the building.
LiDAR assist for the hard landscaping
Separate short LiDAR captures (5 m depth range) of the patio, steps, boundary walls, side passage, shed — anywhere you might later want real measurements for a landscaping design. LiDAR gives you the accurate ground geometry; photo mode gives you the look.
Exterior orbit & the cool extras
The captures that turn this from a scanning chore into a capability library.
House exterior (photo mode)
- Slow orbit of the whole house at whatever access allows — 3 heights where possible, 100–200 photos, corners covered from two angles.
- Step well back for a few frames per side so the roofline makes it in.
- This becomes the building-dissection department's first real-geometry calibration target: the photo-to-massing pipeline's output checked against an actual measured mesh of the same building.
Furniture object library (photo mode, masking on)
Pick 3–5 hero pieces — the ones that would appear in any future restyle render. Orbit each at three heights, 60–100 photos, masking ON. Result: real 3D props to drop into future room designs instead of image-gen approximations.
The AR party trick (free forever)
Export every room mesh as USDZ as well as GLB. USDZ files open natively on iPhone in AR — point the phone at the floor and walk around a 1:1 or miniature copy of your own rooms. Zero subscription, zero viewer needed; it's built into iOS.
The 3D-print miniature
Export the whole-floor walks (or the exterior capture) as STL. Any print bureau — or a friend with a printer — can produce a physical miniature of the house. Gimmick? Partly. But it's also the single most persuasive demo of the spatial lane that exists: here is a physical object made from a phone walk-around.
In-app measurements while you still can
Polycam lets you drop measurements on any capture in-app. On day 4, take screenshots of key measurements (room widths, ceiling heights, garden lengths) as a belt-and-braces record alongside the CSV exports.
Export discipline
The subscription's real product is the export set. Same-evening exports, every format, filed to the same tree as the existing properties. Assume the cloud copy dies when the trial does.
| Capture type | Export formats | Why each |
|---|---|---|
| Room mesh (LiDAR) | GLB + OBJ + USDZ | GLB = our three.js viewers · OBJ = universal interop · USDZ = iPhone AR |
| Room mesh (LiDAR) | PLY | Point cloud — the raw measurement record |
| Floor plan (Room mode) | CSV + DXF + PDF + PNG + SVG | CSV = dimensions data · DXF = CAD · PDF/PNG/SVG = human-readable |
| Merged floor plan | Same five, per floor | The un-reconstructable asset — do not skip |
| Garden / exterior (photo) | GLB + OBJ + PLY | Viewer + interop + raw cloud |
| Hero rooms (splat) | PLY / SPLAT | Radiance field — locked behind sub after cancel |
| Whole-floor walks | GLB + STL | Explorer model + 3D-print miniature |
| Every room | 360 pano image | Insurance-grade dated visual record |
Filing tree
ground-living-mesh, ff-bed2-plan) — attribution by filename beats memory three days later.Tape verification
One tape measurement per zone turns "scan values" into "verified dimensions" — the same protocol that upgraded the games room from UNVERIFIED to VERIFIED.
- Ground floor: one long uninterrupted wall (skirting height, corner to corner) — compare against the merged plan's CSV.
- First floor: same, one long wall. Plus one ceiling height per floor.
- Each garden: one straight boundary edge or patio length.
- Pass mark: scan within ±2 cm of tape → mark that zone VERIFIED in
DIMENSIONS.md. Outside tolerance → note the delta, keep the tape value as truth.
What we build after — the payoff list
Everything below runs on the exported files and local tooling. None of it needs Polycam again.
| Build | What it is | Runs on |
|---|---|---|
| Per-room 3D viewers | Drag-to-spin twins of every room, same as the head office pair | GLB + three.js (local) |
| House explorer | Floor-explorer pattern: pick a floor, pick a room, spin it — the whole home as one navigable dashboard property | GLB + floor-walk models |
| Restyle lane, any room, forever | Any room can enter the guide + gate restyle pipeline at any future date — the twin already exists | Mesh + survey photos |
| Garden design lane | Landscaping restyles on measured hard bones — patios, planting schemes, garden rooms | Garden meshes + LiDAR assists |
| Insurance archive | Dated 360 + mesh record of every room and its contents — best-possible claims evidence | 360 panos + meshes, filed |
| Measured-survey substitute | Tape-verified plans ≈ the €500–1,500 survey an architect orders before extension work | Merged DXF/CSV plans |
| Furniture prop library | Real pieces as 3D props in future designs | Object GLBs |
| AR walkthroughs | Rooms at 1:1 or dollhouse scale on any iPhone, natively | USDZ |
| Printed miniature | Physical model of the house — the spatial lane's best demo object | STL |
| Dissection calibration | Photo-to-massing pipeline scored against real measured geometry of the same building | Exterior orbit mesh |
Cancel-day checklist
Only after every line passes.
- Every capture exported in every format in the section 7 table — audited against the filing tree, file sizes sane, GLBs open in a local viewer.
- Merged floor plans exported per floor (the un-reconstructable asset).
- Splats exported — these are the most likely to be locked behind the paywall after cancellation.
- Measurement screenshots taken for key dimensions.
- Tape checks done,
DIMENSIONS.mdstatuses set. - Then cancel — in-app / App Store subscription page, screenshot the confirmation and the access-until date.