3D Gaussian Splatting Without the Python Headache
If you've ever tried to get into 3D Gaussian Splatting, you know the drill: install Python, wrestle with PyTorch and the right CUDA version, then set up COLMAP separately and hope the whole stack cooperates with your GPU. It's a lot of yak-shaving before you see a single splat. Spirula Studio takes a different approach—it packages the entire pipeline, from raw photos or video to splat to textured mesh, into one self-contained binary.
What It Does
Spirula Studio is an open-source application that trains 3D Gaussian Splatting models. You feed it photos or video, and it handles the whole chain: structure-from-motion, training, and export. It runs on NVIDIA, AMD, Intel, and Apple GPUs via Vulkan compute, which means you're not locked into one vendor's ecosystem.
The architecture is what sets it apart. Instead of relying on Python and PyTorch, it compiles down to a single binary you download, unzip, and double-click. There's a GUI for interactive work and a CLI for remote or cloud GPU training—by default, the spirula train command serves a viewer over HTTP, so you can forward the port over SSH and watch training progress in your browser. The project is GPLv3 licensed and ships binaries for Windows, Linux, and macOS.
Why It's Cool
The dependency story is the headline. No Python, no PyTorch, no separate COLMAP install. For anyone who's spent an afternoon debugging CUDA toolkit mismatches, that's a genuine relief. The Vulkan backend means the same binary works across GPU vendors, and the README notes it's the most tested and produces a smaller build than the CUDA backend.
Memory efficiency is genuinely impressive. Quantized training lets you fit up to 10 million full-SH3 Gaussians in 8 GB of VRAM. That's a meaningful number—it means you can train substantial scenes on consumer hardware rather than renting a beefy cloud instance.
It handles cameras most tools punt on. Native support for 360° and equirectangular cameras means you can load a dataset and train directly, no undistortion step. Fisheye support is there too.
The training strategy is a hybrid. It combines advantages of MCMC, IGS+, and MRNF into one approach, aiming for sharper results and fewer floaters across everything from small objects to large scenes.
The extras are practical, not decorative. Built-in fast SfM, AI masking, and frame extraction from video mean you're not stitching together separate scripts. Modified bilateral grid and PPISP handle exposure and white balance correction without the color shift or darkening you sometimes get elsewhere. There's also depth/normal output, meshing, skybox, and linear color support.
Recent updates suggest active development. Metric scale recovery from telemetry metadata in video and image formats addresses the classic "my reconstruction is too big or tilted" complaint. LoMa feature support helps with difficult datasets, and macOS/Apple Silicon training is now validated. The GUI and CLI are multilingual across a dozen languages.
How to Try It
- Head to the Releases page and grab the binary for your platform (Windows, Linux, or macOS).
- Unzip it and double-click to open the GUI.
- Load your photos or video and start training.
If you're on a remote or cloud GPU, use the CLI instead:
spirula --help
spirula train
The train command serves a viewer on an HTTP port you can forward over SSH.
Prefer to build it yourself? The repo supports two backends—Vulkan (recommended, cross-platform, smaller binary) and CUDA. Full build instructions are in the repository.
Final Thoughts
Spirula Studio is aimed squarely at people who want to do Gaussian Splatting without becoming a Python environment manager first. If you've got a Vulkan-capable GPU and a folder of photos, this is about as low-friction as the pipeline gets. The 8 GB VRAM ceiling for 10M Gaussians and the native 360° support make it worth a look even if you've already got a working setup elsewhere. It's a young project with a fast-moving changelog, so expect rough edges—but the core idea, one binary that does the whole job, is exactly the kind of simplification this space needed.
Follow @githubprojects for more developer tools and open source projects.