RealityScan’s march toward deeper automation took a decisive step forward this week with the launch of version 2.1. While the major 2.0 milestone earlier this year focused on alignment quality, AI masking, and LiDAR ingestion, RealityScan 2.1 reaches further into the pipeline, formalizing command line processing, enabling remote execution, and widening support for SLAM and classified point clouds.
The throughline of the release is automation. The team published full CLI documentation, now paired with Python examples on the Epic Developer Community, making the command interface far easier to adopt for users integrating RealityScan with Gaussian Splatting, COLMAP based pipelines, or custom cloud renderers. Just as notably, Epic introduced a new Remote Command Plugin that exposes RealityScan over gRPC or REST. With this in place, RealityScan can now run as a controllable headless service on local networks or in the cloud. Projects can be queued, monitored, and handed off between machines without ever opening the UI.
Alongside the new API layer comes a rebuilt templating system, a modernization that allows teams to output structured project metadata in JSON and generate custom reports or even their own export formats. For radiance field practitioners working with version controlled datasets, or for studios managing multi stage capture to splat workflows, the ability to extract consistent, machine-readable representations of a project’s structure is a quiet but meaningful upgrade.
The update also widens support for LiDAR and SLAM data in ways that directly benefit hybrid pipelines. Handheld SLAM cameras, including trajectories from systems like XGRIDS, can now be imported alongside traditional imagery. RealityScan can fuse these paths with photogrammetry or aerial scans, then generate virtual cameras using available priors. For city scale or multi rig environments, the ability to anchor training to SLAM or IMU paths dramatically reduces drift and stabilizes radiance field optimization. Classification from LAS and LAZ files is now recognized as well, enabling selective meshing or class-aware pruning before model training.
For aerial and vehicle based pipelines, RealityScan’s trajectory importer has been rebuilt and renamed. The new system supports custom log definitions, flexible Euler angle orders, and updated mount direction handling across yaw, pitch, and roll. The underlying registration exports were also updated, bringing new support for OpenCV and XMP outputs and improved COLMAP exports that correctly reference original distorted images.
RealityScan 2.1 also extends the rendering and texturing surface of the application. Models can now be rendered from the exact intrinsics and extrinsics of any input camera, including undistorted variants, enabling tighter supervision for radiance field training or dataset verification. Normal maps can be rendered in camera or world space with conventional orientation. UV unwrapping has been cleaned up with clearer visualization, more aggressive large-triangle filtering, and texture defragmentation tools activated by default. As with every RealityScan release, the texturing engine received more fixes than additions, addressing crashes, hangs, memory exhaustion, and various inconsistencies that disproportionately affected high polygon or large area scenes.
This new version also brings a command line build of RealityScan for Linux. It runs inside a bundled Wine environment, but it exposes full CLI workflows for alignment, reconstruction, texturing, and export. Epic is clear that the desktop UI should not be used on Linux due to graphical and focus issues. For studios already running Gaussian Splatting training on Linux GPUs, being able to colocate reconstruction and training on the same hardware should be nice.
RealityScan 2.1 is available now through the Epic Games Launcher, and the new CLI documentation, Python samples, and API materials are live on the Epic Developer Community. Read the full release notes here.
