Appearance
Complete Camera Alignment with RealityScan
Step 0 | Download and Install RealityScan
This tutorial uses the desktop version of RealityScan. Epic Games provides installation and version management through the Epic Games Launcher, so you first need to install the Launcher and have an Epic Games account you can sign in to.
Official downloads and documentation: RealityScan download page | Official installation guide
Download Epic Games Launcher: Open the RealityScan download page and click Download Launcher. After the download finishes, run the installer and follow the setup wizard. If the Launcher is already installed on your computer, continue to the next item.
Sign in to your Epic Games account: Start Epic Games Launcher and sign in with an existing Epic Games account, or create a new account when prompted. RealityScan cannot be installed while you are signed out.
Find RealityScan in the Launcher: Click Unreal Engine in the left navigation bar, then open the RealityScan tab. Choose the version you want to install; for first-time use, the latest stable version selected by default is recommended.
Install the software: Click Install, review and accept the pricing plan and End User License Agreement shown in the interface. If needed, click Browse to choose the installation location, then click Install again to start downloading and installing.
Launch RealityScan: After installation, click Launch on the RealityScan tab. Once the main interface opens successfully, continue to Step 1 to create a project and import your source media.
Prepare Your Data
Capture coverage: Keep sufficient overlap between adjacent views. Move continuously around the subject and avoid large jumps in viewpoint.
Image quality: Minimize motion blur, reflections, large textureless areas, and abrupt exposure changes. Avoid frequent zooming while recording video.
File management: Decide on the project directory in advance. Do not move image files casually after export; the image paths recorded in the JSON must continue to point to the frame files correctly.
Video frame extraction: The screenshots in this guide use Jumps' length(s) = 0.500000, which extracts one frame every 0.5 seconds. A shorter interval produces more frames and increases computation time; for faster-moving scenes, reduce the interval as needed.
Step 1 | Open RealityScan and Start with an Empty Project
After launching RealityScan, open the WORKFLOW page. The Inputs panel on the left is empty, the center shows the 3D view, and the top toolbar includes entries such as Images, Folder, Align Images, and Export.
Figure 1. Initial interface of a new RealityScan project
Checkpoint: Confirm that the current project contains no leftover inputs or old components. Saving the project file now is recommended so you can recover your work if needed.
Step 2 | Import a Video or Image Folder
You can drag a video file or image folder directly into RealityScan, or use the Images / Folder buttons on the top toolbar. When you drag in a video, the Import Video Sequence dialog appears.
Figure 2. Video import and frame extraction settings
Recommended Settings
After choosing “create a frame folder next to the video� as the save location, RealityScan creates a folder beside the source video and stores extracted frames such as frame000001.png in it.
This folder is also required in Step 10 when you upload the 3DGS training data. Keep it intact; do not delete, rename, or move it casually. You can start with Jumps' length(s) = 0.5 seconds.
If you are importing a photo folder that has already been prepared, simply keep that original photo folder.
Action: After confirming the settings, click OK. Do not rename or move the source video or frame directory while the import is in progress.
Step 3 | Wait for the Video to Be Split into Images
An Importing Video progress indicator appears. Processing time depends on video duration, resolution, frame extraction interval, and disk speed. You may minimize the progress panel, but do not terminate the application.
Figure 3. RealityScan importing the video and generating frames
Checkpoint: When processing is complete, the Inputs panel on the left should show a continuous sequence of frame images. If the number of frames is unexpectedly low, check whether the video is damaged or the extraction interval is too large.
Step 4 | Check the Extracted Frames and Image Quality
In the example, 121 images are imported with filenames such as frame000001.png and frame000002.png. Select any frame and inspect sharpness, exposure, and subject coverage in the 2D view. At this stage, Registered will usually still show No because alignment has not yet been performed.
Figure 4. 121 video frames imported and waiting for camera alignment
Step 5 | Click Align Images to Start Camera Alignment
In the WORKFLOW area at the top, locate Align Images. This step detects features, matches images, estimates camera poses, and generates sparse scene points at the same time.
Figure 5. Click Align Images on the top toolbar
Step 6 | Wait for Alignment to Finish and Check Components
Wait until the Aligning Images process finishes. More images and higher resolutions require more processing time. After alignment, the 3D view should show the camera trajectory together with a sparse point cloud.
Figure 6. RealityScan performing image alignment
Success criteria: Most images should show Registered, the camera distribution in the 3D view should be continuous, and the sparse points should outline the target scene. Handling multiple components: If multiple Components appear in the left panel, the image groups were not fully connected during alignment. Prioritize the component with the largest number of registered cameras and a correct point cloud.
Step 7 | Select the Valid Component and Open Export
The example produces multiple components: Component 0 has 12/121 cams, Component 1 has 107/121 cams, and Component 2 has 2/121 cams. Normally, select the largest geometrically correct component - Component 1 in this example -and then click Export in the [3] Output section on the top toolbar.
Figure 7. Select the valid component and click Export
Do not overlook component selection: The JSON and PLY files must be exported from the same valid component. If they are exported from different components, their coordinate systems, camera counts, and point-cloud positions may not match.
Step 8 | Export the Camera Transformation JSON
In the Export Your Creation window, go to the Camera Alignment section and select Radiance Fields Transformation File (JSON format). Choose a folder to save the exported JSON file and image files.
Figure 8. Select Radiance Fields Transformation File
Figure 8.1 Make sure to export the images together with the JSON file
Figure 8.2 After a successful export, the images and JSON file will be saved in the same folder
When exporting the Radiance Fields Transformation File, make sure to enable image export and select a folder for the exported images. The filenames of all exported images must exactly match the corresponding image names referenced in the Radiance Fields Transformation File. Do not rename any of the exported images.
Step 9 | Export the Sparse Point Cloud PLY
Click Export again, scroll to the Sparse Scene Structure section, and select Sparse point cloud as Polygon File Format (*.ply). Save it as sparse_points.ply, or use the filename required by your downstream tool.
Figure 9. Select Sparse point cloud as Polygon File Format (*.ply)
Step 10 | Upload the Images, JSON, and PLY and Submit 3DGS Training
Open the Create 3DGS from images/video window, enter a Project name, and select Photos under Source Type. Then upload the following three training assets generated in the previous steps, all from the same RealityScan alignment component. Image folder: In the Photos section, click Select folder and select the image files exported in Step 8. JSON file: Expand Advanced Training Files. Under Camera poses, click Select .json and upload the transforms.json exported in Step 8. PLY file: Under Initial sparse point cloud, click Select .ply and upload the sparse_points.ply exported in Step 9.
Figure 10. Upload the image folder, camera-pose JSON, and initial sparse point-cloud PLY in the 3DGS creation window
Submit training: Confirm that all three files are shown as selected and verify that the project name is correct. Then click Start reconstruction in the lower-right corner to submit the 3DGS reconstruction and training job.