Train on a native core
Run 3DGS pipelines on a C++23 and CUDA foundation with save-and-resume built into the same desktop workflow.
Read the docs
0%
3D preview unavailable
Click scene Jump to the next camera angle
LichtFeld Studio
Open-source 3D Gaussian Splatting, end to end
Train, edit, and render photorealistic 3D scenes from photos or video — locally on your GPU, with Python plugins and MCP automation built in.
Native 3DGS trainingSplat editingPython + MCPWindows & LinuxGPLv3
Product tour
One pass through the current build: bring in a scan, train it, inspect and edit the result, then export it — all in the native workspace.
Workflow
LichtFeld Studio brings training runs, live inspection, scene editing, plugins, and MCP control into one desktop workspace built around 3D Gaussian Splatting.
The same runtime handles compute, scene state, and tool extensions, so teams can work inside one consistent environment instead of maintaining a disconnected stack.
The same application handles training output, scene editing, and live inspection.
Edit in context
Brush, lasso, polygon, crop, align, and compose splats while working against the live scene.
See scenes in the galleryAutomate the workflow
Use Python plugins and MCP endpoints to build custom panels, operators, batch tools, and external workflows.
Browse plugins
Session 2: How the Grace Cathedral Splat Was Made
Session 2 is on YouTube. Vincent Woo tells the story of the Grace Cathedral splat, from a Frankenstein camera rig and a crashed drone to 10 million Gaussians in the browser, plus a first look at his next project.
Session 1: Faster Training, a LiDAR for the Lab, and What You Asked For
Our first Session is live on YouTube. In about 37 minutes we cover a 3x training speed-up, your survey answers, a LiDAR scanner for the lab, and the new .licht project format.
3D Gaussian Splatting
3D Gaussian Splatting (3DGS) turns ordinary photos and video into photorealistic 3D scenes that render in real time. LichtFeld Studio is a free, open-source Gaussian Splatting application that covers the whole workflow — training, editing, rendering, and automation — locally on your own GPU.
Instead of meshes or NeRF-style neural networks, a Gaussian Splat scene is built from millions of 3D Gaussians that rasterize in real time. That speed makes 3DGS a natural fit for scanning real places and objects, drone and aerial capture, virtual production, and interactive visualization.
- What is 3D Gaussian Splatting (3DGS)?
- 3D Gaussian Splatting is a technique for reconstructing photorealistic 3D scenes from ordinary photos or video. The scene is represented as millions of 3D Gaussians — splats — that rasterize in real time, making 3DGS a fast alternative to NeRF for radiance field rendering.
- Is LichtFeld Studio free?
- LichtFeld Studio is open source under the GPLv3 license, so you can always build it for free from source on GitHub for Windows and Linux. The prebuilt Windows binary is available through the LichtFeld portal for a minimum contribution of $30, which funds ongoing development.
- What hardware do I need to train Gaussian Splats?
- Training runs locally on an NVIDIA GPU with compute capability 7.5 or newer and at least 8 GB of VRAM. The core is written in C++23 and CUDA for fast 3DGS training and real-time rendering.
- Can I edit and automate Gaussian Splat scenes?
- Yes. LichtFeld Studio includes a full splat editing suite — brush, lasso, and polygon selection, cropping, alignment, and composition — plus Python plugins and MCP endpoints for automating 3DGS workflows.
Community
LichtFeld Studio is developed in public by 56 contributors under GPLv3. Development discussion, support, and scene sharing happen on Discord, and the contribution backlog is open to everyone.
Sponsors
Foundational Sponsor
Gold
Get LichtFeld Studio
Download the Windows build, or compile from source for Windows and Linux. The docs cover hardware requirements and your first training run.