9.2 Single Scene Processing
Taking one continuous scan from capture to a finished model: deciding how long the scene can be, importing it, previewing the data before you commit, setting the few parameters that matter, and running the reconstruction.
When to Use Single Model
Single Model turns one continuous scan into one model. It is the default and the starting point for most jobs: any scene a single scanning session can cover. If the site needs more than one session, or a drone, you are in Map Fusion or Aerial-Ground territory instead, covered on 9.3 and 9.4.
Device support. Single Model works on the L3, L2 Pro, K2, and PortalCam. What differs between them is how long one scene can run, which Section 2 works through.
Decide How Long the Scene Can Be
Plan a single scene at 12 to 15 minutes and check it against your workstation before you commit. Nothing in LCC Studio stops you from going longer. That is the problem: the software has no time limit and no timer, so it will accept a scene it cannot finish and fail hours into the run.
Two ceilings apply, and they come from different places. One is the scanner. The other is the machine that will reconstruct the data, and that is almost always the one that binds.
The scanner ceiling
| Device | Maximum single capture | Set by |
|---|---|---|
| L3 | Not yet published | Not yet published |
| L2 Pro | 90 minutes | Firmware. Recording stops at the limit. |
| K2 | 90 minutes | Firmware. Recording stops at the limit. |
| PortalCam | 60 minutes | Firmware. XGRIDS additionally advises against exceeding 30 minutes in any mode. |
The PortalCam has a working limit well below its ceiling. XGRIDS does not recommend PortalCam captures over 30 minutes, for single-scene reconstruction or for fusion segments. Longer captures tend to produce drift and layering artifacts in the result, and layering is not correctable in post. Treat 30 minutes as the hard planning number on a PortalCam job regardless of what the battery or the firmware allows.
The workstation ceiling
Reconstruction memory demand climbs at roughly 2 GB per minute of capture data. That single rate produces both figures below, which is why they sit where they do.
| Installed memory | Reconstructs reliably | Real risk of failure above |
|---|---|---|
| 64 GB | Up to 30 minutes of capture | 45 minutes |
| 128 GB | Up to 60 minutes of capture | 90 minutes |
These are ceilings, not plans. A failed reconstruction cannot be partially recovered, and reconstruction runs at roughly 20 minutes of processing per minute of capture, so a 45-minute scene is most of a day of compute before you learn whether it worked. Planning at 12 to 15 minutes leaves room for unexpected coverage without approaching either number.
XGRIDS recommends against capturing a full-length dataset in one session, on any hardware. A fault anywhere inside a long capture, whether SLAM drift, environmental interference, or a battery problem, makes the whole dataset difficult to recover and may force a complete rescan. The recommended approach for a large scene is to split it into shorter segments and combine them with Map Fusion, which is both safer and more flexible. Field procedure for the split is on 6.2 Scan Splitting.
If the job genuinely needs one long scene
Some scenes cannot be split usefully, and the answer is hardware rather than technique. For a single capture in the 60 to 80 minute range, XGRIDS specifies 128 GB of RAM as the minimum, a GPU with 24 GB of VRAM in the RTX 4090 class or better for headroom on a model that size, and at least 240 GB of free SSD space. Even at that specification an 80-minute single scene sits at the upper boundary of what will reconstruct, and the risk of losing the whole capture to a single field fault does not go away. Split it if you can.
The software tells you before you commit, and again when it fails. LCC Studio estimates memory demand from the uploaded data and displays it. A red Estimated Memory reading means the scan is too large for the machine, and that is the moment to change something: split the capture, add memory, or drop Reconstruction Quality to Standard. Commit past a red estimate and the run does not degrade gracefully. It exhausts system memory and fails, typically reporting error 0x34040060, with nothing to recover.
Capture One Continuous Scene
Scan the space in one unbroken session using the field technique for your device in Modules 3 through 7. Three things matter specifically for a clean Single Model reconstruction:
- Keep it one continuous capture. Do not power off or break the scene into pieces. A single model comes from a single capture; multiple pieces belong in Map Fusion, and no XGRIDS scanner can resume a scan from a breakpoint.
- Close your loops. Return to the start and re-walk shared ground so the SLAM solution closes cleanly. Open loops show up later as drift and doubled surfaces that no setting fully fixes.
- Wait out the write process before touching the device. Stopping the scan does not end the job. The scanner writes the project file to internal storage, which takes 30 to 60 seconds and longer after a long capture, and the green indicator flashes rapidly the whole time. Powering off before it goes solid green corrupts the file and the capture is not recoverable.
If the scene is outdoors and needs real-world coordinates, capture it with RTK so the model can be georeferenced during reconstruction. If it is indoors and you want a floor plan, plan to enable Spatial Recognition at reconstruction, which is an L3, L2 Pro, and PortalCam feature. Only Single Model projects offer it, so a floor-plan job has to fit one continuous scan. Neither can be added after the scan.
Import and Preview the Data
- Create the project. Click Create and select Single Model. Connect the device by USB and let LCC detect and import the capture, or point it at the capture project folder. Import the top-level project folder, not a sub-folder, and keep the path free of spaces.
- Confirm the device type. LCC reads the data and displays the device it detects. Check it matches the scanner you used before going further.
- Check the memory estimate. Before anything else, read the Estimated Memory indicator against the tiers in Section 2. Red means this machine cannot hold this scene, and every minute spent past that point is wasted.
- Preview the point cloud before committing. Click Point Cloud Preview. LCC checks the capture and shows the trajectory and a rough cloud, flagging problems while you can still fix them, before you spend hours on a reconstruction that was never going to succeed.
Set the Parameters That Matter for a Single Scene
These are chosen at project creation and lock at reconstruction. Confirm them before you start, because changing one later means reprocessing. The full settings and advanced options are on 9.8 LCC Studio Tools.
| Setting | What to do |
|---|---|
| Reconstruction Quality | Standard for client work. Slow for final, detail-critical models. Fast for a quick preview only. Dropping to Standard is also the first move when memory is tight. |
| Maximum Gaussian Points | Keep at or below 25 million on a consumer GPU. In Single Model this caps the model's total points and must fit GPU memory. |
| Coordinate System Conversion | Leave on None indoors. For an outdoor RTK scene that must align to GIS or BIM, pick the target system. It decides automatically: valid RTK converts, no RTK builds in local coordinates. |
| Low-Memory Mode | Enable only if a reconstruction fails for lack of RAM. It writes intermediate data to disk and extends processing time. Available in Single Model only. |
| Spatial Recognition | Builds the structured floor plan the 3D Layout and Scene Report tools use, and prepares indoor structure for BIM and Revit work. Enable it at reconstruction if the job needs a floor plan (9.11) or the Revit workflow (9.6); those are built from it and cannot be produced without it. Single Model only. L3, L2 Pro, and PortalCam, indoor, needs more than 8 GB of GPU memory. See 9.11 Floor Plans and 9.6 LCC Revit Plugin. |
| HD Enhancement | Enable if you shot supplementary close-up photos of key areas. See 9.5 HD Enhancement. |
Run the Reconstruction
- Start it. Click Start Reconstruction and confirm. LCC processes the scene automatically and shows a progress ring and stage bar on the model card.
- Leave LCC open. Do not close LCC Studio while the model is generating. Closing it interrupts the job; the card then shows a failure, and Continue Generation or Restart Generation on the card menu re-queues it.
- Keep disk space free. Reserve at least twice the capture size on an SSD separate from the install directory, or the run can fail partway for lack of room.
- Do not compete for the GPU. Other GPU-heavy work on the same machine during a run can starve the reconstruction and fail it.
If a reconstruction fails for memory, the capture is not lost. Drop to Standard quality or enable Low-Memory Mode and re-run. If it fails again, the scene is longer than this machine can hold and the honest options are more memory or a split capture, not another attempt with the same settings. Memory tiers are in Section 2; the full hardware tables are on 10.7 Device Specifications.
After Reconstruction
When the card switches to a thumbnail, the model is ready. From here the job moves to editing and delivery:
- Navigate and check it in the viewer. Set a comfortable camera speed and walk the model for gaps or drift. See 9.8 LCC Studio Tools.
- Edit it. Scene-level cleanup, measurement, viewpoints, and publishing are in 9.9 Scene Editor Tools. Point-level cropping and color grading are in 9.10 Model Editor Tools.
- Export or publish. Choose a delivery format or publish a web link on 9.7 Export Formats.
©2026 Alpine Reality Capture LLC • XGRIDS Pro Guide™ • Site Disclaimer