One App to Train, Edit, and Export 3D Gaussian Splatting Scenes
If you've worked with 3D Gaussian Splatting, you know the drill: one tool to train, another to view, a script to export, and a pile of glue code holding it all together. LichtFeld Studio is an attempt to collapse that workflow into a single native application—train, inspect, edit, and export without stitching together separate tools.
What It Does
LichtFeld Studio is a modular workstation for 3D Gaussian Splatting. It's a native desktop app that handles the whole pipeline: training new scenes from COLMAP datasets, resuming from checkpoints, inspecting reconstructions in real time, editing gaussian selections, and exporting results. The project describes itself as built for users who need more than a training script or a standalone viewer, and the feature list backs that up.
Beyond the core training and viewing loop, the app supports Python plugins with plugin-local dependencies, and automation through MCP (Model Context Protocol) resources and tools plus embedded Python. Exports come in several formats—PLY, SOG, SPZ, or a standalone HTML viewer—so you can hand off results in whatever shape the next step of your pipeline needs. Undo and redo are built into the editing workflow, which matters more than it sounds once you're manipulating gaussian subsets and scene nodes.
Why It's Cool
-
The workflow consolidation is the real pitch. Training, visualization, editing, and export living in one app means no more shuttling files between a training repo, a viewer, and a custom export script. For anyone who's done this the hard way, that's a meaningful reduction in friction.
-
Editing gaussians directly. Selecting, transforming, and editing subsets of gaussians with undo/redo support turns the reconstruction into something you can actually work with, not just look at. That's a different posture than most 3DGS tooling, which tends to stop at "here's your trained model."
-
Automation isn't an afterthought. MCP resources and tools plus embedded Python suggest the team is thinking about LichtFeld as a component in larger pipelines, not just a GUI. Tool builders can drive it programmatically; researchers can script experiments around it.
-
Multiple export formats. PLY for interoperability, SPZ for compactness, SOG, and a standalone HTML viewer for sharing. The HTML export in particular is a nice touch—you can ship a scene to someone who doesn't have any 3D software installed.
-
A checkpoint-friendly training loop. Resuming from checkpoints is called out explicitly, which tells you the app is designed for iteration—the reality of research, where you tweak and rerun rather than train once and walk away.
How to Try It
The easiest path is the Windows download via the project's portal:
- Grab the Windows build from portal.lichtfeld.io.
- If you'd rather build it yourself, follow the build from source guide in the repo.
- Once running, load a COLMAP dataset to start a training run, or open an existing checkpoint to inspect a reconstruction.
- If you want to extend the app, the Python API and plugins docs cover plugin development, and the MCP guide covers automation.
The repository is at github.com/mrnerf/lichtfeld-studio. There's also a YouTube walkthrough showing the app's capabilities, which is worth a look before you commit to a build—it'll give you a sense of whether the workflow matches how you actually work.
Final Thoughts
LichtFeld Studio is aimed at people who've outgrown the "train and pray" phase of 3DGS work: researchers iterating on reconstruction quality, production teams who need to inspect and edit scenes without juggling tools, and developers integrating 3DGS into larger pipelines. If you're just training your first splat, a simpler script might serve you fine. If you're tired of the toolchain sprawl, this is worth a serious look—and the plugin and MCP hooks suggest it'll keep growing in the directions power users care about.
Follow @githubprojects for more developer tools and open source projects.