SplatLoomYour first Gaussian splat, from photos to a scene
A practical local workflow for capturing, generating, inspecting and exporting a Gaussian scene in SplatLoom.
Start with one still subject
Choose a textured, stationary object or a small part of a room. Avoid glass, mirrors and moving people for your first attempt. Gaussian splatting reconstructs views of a real scene; it does not turn a single photograph into an accurate solid model.
Capture overlapping viewpoints
Take roughly 30–150 clear photos as a starting range, with plenty of overlap. Move around the subject instead of standing in one spot and rotating the camera. Keep your distance and exposure reasonably consistent. There is no photo count that guarantees a good result.
Let the default workflow run
Open the Mac preview and select the photo set or a slowly captured video. Start with the recommended settings. SplatLoom prepares the images, estimates camera positions and trains a Gaussian scene on your computer. Processing time depends on the dataset, hardware and available memory.
Inspect before exporting
Rotate the result and look at the areas you actually photographed. Holes at the back of an object often indicate missing coverage. If reconstruction fails, read the processing details and improve your capture before simply increasing the training settings.
Choose an output for the job
Save PLY for a compatible Gaussian viewer or editor, PNG for a still image, MP4 for an orbit video, or a self-contained HTML file for interactive local viewing. A Gaussian PLY is not a mesh for 3D printing. The HTML file may be large and must be hosted separately if you want a public link.