Splat Local

Walk through a space once. Get a 3D scene you can fly through forever.

Turn a video walkthrough into a 3D Gaussian Splat and watch the scene resolve out of the fog, live, in your browser — no cloud, no CUDA, nothing leaves your Mac. Downloads as .ply (plus .spz / .sog when available).

🎥 record a walkthrough	✨ get an explorable 3D scene

_{An 11 s phone-style walkthrough → a splat you can fly through with WASD/arrow keys. Reconstructed end-to-end in ~25 min on an M5 Pro MacBook — 165 frames, COLMAP poses, 30k training steps. Footage: Pexels #7578547 (free license).}

Why

• Actually local. Poses, training, and the viewer all run on your machine — nothing uploaded, no API keys, no CUDA required.
• You watch it build. Training checkpoints stream straight into the browser viewer, so the scene sharpens from fog into a real space in real time instead of a progress bar.
• Quality that holds up. COLMAP-grade poses + a Metal-native trainer that matches CUDA gsplat output, not a lightweight approximation.

How it works

video ──▶ sharp frames ──▶ camera poses ──▶ splat training ──▶ cleanup/export
          (ffmpeg +        (COLMAP, or       (Brush: Metal-      (splat-transform:
           sharp-frames)    Depth Anything 3  native 3DGS w/      floater removal,
                            on MPS)           MCMC + mip AA)      .ply/.spz/.sog)

• Poses: COLMAP (pycolmap) with sequential matching + loop detection — best quality. Optional experimental backend: Depth Anything 3 running on Apple's MPS — much faster, slightly lower fidelity.
• Training: Brush — a Rust/Metal Gaussian-splat trainer that matches CUDA gsplat quality (MCMC densification, Mip-Splatting antialiasing, optional LPIPS loss). It exports .ply checkpoints throughout training, which the UI streams into a live Spark viewer.
• Everything runs on your Mac. No cloud, no CUDA.

Quickstart

./setup.sh        # installs ffmpeg/uv if missing, syncs Python env, fetches/builds Brush
./run.sh          # serves http://127.0.0.1:8000

Upload a video, pick a preset, watch it build. Presets:

Preset	Frames	Steps	Approx. time (M-series Pro)
Preview	100	10k	~20–30 min
High	200	30k	~2–3 h
Max	250	45k	~4 h

Capture tips (quality lives and dies here)

• Move slowly in an orbit/arc with lots of overlap; end near where you started (loop closure).
• Lock exposure/white balance if you can; 4K 60 fps gives the frame picker more sharp frames.
• Avoid moving subjects, whip pans, and textureless walls/sky-only shots.

Notes

• Optional DA3 pose backend: uv sync --group da3 (Python 3.12 venv, installs PyTorch). Uses depth-anything/DA3-LARGE by default; override with DA3_MODEL=depth-anything/DA3-SMALL ./run.sh for speed.
• Optional .spz/.sog export + floater cleanup uses npx @playcanvas/splat-transform (needs Node).
• Why not LingBot-World? It's an image→video world generator (28B params, CUDA-only, no 3D output) — the wrong tool for video→3D reconstruction, and it can't run on a Mac. This project uses the reconstruction stack that modern world-model papers themselves use for geometry.

Layout

server/ FastAPI + pipeline stages · web/ vanilla-JS UI + Spark viewer · vendor/ Brush binary + viewer libs · jobs/ per-run work dirs (gitignored) · docs/api.md API contract