Xangle
Xangle Portal
Sign in

Product

Xangle Camera Server

Create & reuse camera references

Registration & alignment

Build one clean reference dataset for the rig, then reuse it until a camera moves. Use Cloud/3D for the fastest managed setup, RealityScan/3D for a Windows desktop 3D workflow with reusable reference files, or Track/2D for a fast fixed-arc fallback with no external tool.

Choose the engine

  • Default. Cloud/3D runs Xangle-hosted 3D camera registration with no local install. Use this when you want managed processing and fast setup.
  • RealityScan/3D is a full desktop 3D camera registration workflow via RealityScan (Epic Games, Windows only). Use this when you want a local 3D workflow and reusable reference files for later processing.
  • Track/2D is an in-app 2D alignment path. No external tool is required, but it is meant for fixed arcs and does not recover real 3D camera poses.

Engine comparison

On clean, properly lit datasets, all three registration engines can produce the same alignment result. For the cleanest alignment, follow the capture guidelines above.

  • Cloud/3D: Efficiency 7/10. No extra app to install. There is no warranty on result quality, but it is generally good in normal capture conditions.
  • RealityScan/3D: Efficiency 9/10. Excellent quality in most setups, but it requires a RealityScan license from Epic Games and runs on Windows only.
  • Track/2D: Efficiency 5/10. Fast and fully in-app. Best for fixed arcs, not full 360 rigs.

Cloud camera registration

  • Cloud camera registration runs on Xangle workers, so no extra software installation is required.
  • Images are uploaded only for the active job and are deleted after results are downloaded.
  • Usage is measured with camera registration credits: one registration job costs one credit, regardless of camera count.
  • Credits are consumed only when the cloud registration job completes successfully.

Install RealityScan

RealityScan is the default 3D alignment engine in Camera Server. Epic's official install flow is:

  1. RealityScan runs on Windows only. Download and install the Epic Games Launcher from the official RealityScan download page.
  2. Open the launcher, create or sign in to your Epic Games account, then open the RealityScan tab under the Unreal Engine section.
  3. Install the latest RealityScan build. In Camera Server, you can usually leave the RealityScan path empty so it auto-detects the install under C:\Program Files\Epic Games.
  4. If RealityScan was installed somewhere else, paste the executable path manually in the Alignment settings.
  5. Run one reference dataset from Camera Alignment or Library to confirm Camera Server can launch RealityScan headlessly and write the reference files.

Cost and licensing

  • Epic currently lists RealityScan as free for students, educators, educational institutions, and individuals or companies under $1 million USD in annual gross revenue.
  • Above the $1 million USD revenue threshold, Epic sells yearly RealityScan seats. The public license page currently lists CA$1,697 per seat per year and also advertises a 30-day trial.
  • Check the official RealityScan licensing page before budgeting or purchasing seats, because public pricing and terms can change.

Learning and support

  • Epic Developer Community provides RealityScan documentation, tutorials, sample datasets, and community forums when you need to learn the tool or troubleshoot the install.

Get a strong camera registration

How to get a strong camera registration

  • Frame the subject in the center so each camera sees the same central mass instead of edge clutter.
  • Use even, stable lighting and avoid flicker, harsh shadows, or blown highlights.
  • Keep the subject still so every camera captures the same moment in time.
  • Prefer textured clothing or surfaces so the alignment engine can match the same 3D environment more reliably.

What RealityScan writes

  • Reference XMP sidecars for each registered camera.
  • registration.csv for the aligned camera rig.
  • A sparse point cloud for the 3D alignment preview.
  • Optional: native COLMAP export for downstream processing when that setting is enabled.

Track/2D alignment

An in-app alignment engine that straightens an arc of photos without RealityScan or any calibration markers. It works directly on the images by tracking the center subject from one camera to the next.

When to use it

  • Best for a fixed camera rig where the cameras do not move between captures. Align once, then reuse the result as the active reference for every later dataset.
  • A fast, fully offline alternative to RealityScan or COLMAP that needs no external tool and no markers.
  • It produces a 2D similarity transform per camera: shift, rotate, and uniform zoom. That is enough to straighten orbit or arc captures, but not enough for full 3D photogrammetry.

Capture requirements

  • Up to 120 degrees of arc. Beyond roughly 120 degrees, adjacent cameras stop sharing enough of the same view and the match can drift. Keep the total arc at 120 degrees or less.
  • A centered subject is required. The engine locks onto whatever sits in the middle of the frame, so the subject must stay roughly centered in every camera.
  • Texture helps. Dress the subject in patterned or detailed clothing when possible. Flat or single-colour surfaces give the matcher less to lock onto.
  • Light it properly. Even, bright lighting gives crisp detail to match. Harsh shadows, blown highlights, or a dim subject reduce the number of usable features.

How it works

  • The first camera is the reference. Each following camera is matched to its neighbor, and those small steps are chained back to the reference to stay in the high-overlap region.
  • Matching runs inside a centered region of interest so the subject, not a busy background, drives the alignment.
  • A common interior crop is computed so every aligned frame stays inside the valid image area with no black edges.

Tuning the settings

  • ROI width and height (%). This is the size of the center region used for matching. Widen it if the subject fills the frame; tighten it if the background is pulling the alignment away from the subject.
  • Max ORB features. Raise it for low-texture scenes to find more matches, but expect slower processing. Lower it when the subject is already richly textured and speed matters more.
  • Sub-pixel refinement iterations. These are the ECC sub-pixel refinement passes. More iterations can sharpen the lock at the cost of time; the default is a good balance.

Reusing the result

  • When the alignment succeeds, the computed transforms are saved and the dataset becomes the active reference.
  • Later datasets captured on the same untouched rig automatically reuse those transforms. If you physically move or re-aim any camera, run the alignment again to refresh the reference.

Important boundary

Track/2D is purely 2D. It does not recover real 3D camera poses, so it is meant for straightening orbit or arc turntable shots, not for full photogrammetry or Gaussian-splat reconstruction.

Set the active reference

  • When a result looks correct, promote that dataset as the active reference from the dataset menu in Library or from the Camera Alignment module itself.
  • Future datasets on the same untouched rig can reuse the active reference automatically, so you do not need to align every capture again.

Recover from a bumped camera

  • If any camera is bumped, re-aimed, re-focused, or moved during the event, capture a fresh alignment shot and promote it as the new reference before continuing.
  • If the rig changes a lot between captures, enable auto re-align for new datasets instead of relying on one long-lived reference.