SplatKit one-click installer 1.0.0

One-click install for ComfyUI (Windows)

SplatKit isn't available through the ComfyUI Manager right now, so this is the easiest way to install the full pack.

How to use

1. Download install_splatkit.bat from the Assets below.
2. Drop it into your ComfyUI folder — the portable folder that holds python_embeded and ComfyUI (where run_nvidia_gpu.bat lives), or inside your ComfyUI folder next to custom_nodes.
3. Double-click it.

It fetches the SplatKit node pack into custom_nodes, installs the node dependencies into ComfyUI's Python, and builds the self-contained CUDA backend (~7.5 GB, Windows + NVIDIA) for the 4D generator and splat trainer.

4D Backend Installer v0.1.0

SplatKit 4D Backend Installer

Installs the Windows CUDA backend that SplatKit's 4DAnyone nodes (Generate Views, Export Frameset, Train Sequence) run in. This is not the ComfyUI-SplatKit node pack. Install that from the ComfyUI Manager or the repo first. The backend is kept out of the node package by design: nothing in the pack downloads or installs binaries at workflow runtime.

Requirements

• Windows x64 with an NVIDIA CUDA GPU
• ComfyUI 0.34 or newer, with the ComfyUI-SplatKit custom node already installed
• ~20 GB free disk for setup (backend ~ 7.5 GB; models are separate, ~ 19 GB)
• Generation has been reported to use ~24 GB VRAM; lower-memory cards are unvalidated

How to install

1. Download SplatKit-4D-Backend-Installer.zip below and extract it — you get installer.bat and install_splat_backend.py.
2. Drop both files into your ComfyUI-SplatKit custom-node folder
   (ComfyUI/custom_nodes/ComfyUI-SplatKit/), or into the ComfyUI root.
3. Run installer.bat. It builds a self-contained Python 3.11 / CUDA 12.8 environment
   under the pack's bin/ folder — downloading uv and a SHA-256-verified gsplat wheel, then
   installing the pinned dependencies. It touches nothing in ComfyUI's own Python.
4. Restart ComfyUI. The Splat Backend Setup node should now report Ready.

Notes

• Linux and macOS backend setup are not supported.

gsplat 1.4.0 · Python 3.11 · PyTorch 2.8 · CUDA 12.8

Prebuilt gsplat 1.4.0 wheel for SplatKit’s isolated Gaussian splatting backend.

• Platform: Windows x64
• Runtime: Python 3.11, PyTorch 2.8.0, CUDA 12.8
• GPU kernels: NVIDIA RTX 20–50 series and selected datacenter GPUs; tested on RTX 5090

SphereSfM CUDA binary v1

Notes:
Prebuilt SphereSfM binary bundle for ComfyUI-SplatKit. You don't need to download this manually — the pack fetches and verifies it automatically on first use (see docs/SPHERESFM.md for offline install).

CUDA build of SphereSfM (COLMAP 3.8 + spherical-camera patches, commit 6b40b2d), used by the SphereSfM dataset nodes for panorama structure-from-motion.

• Platform: Windows x64, Linux x64, CUDA; MacOS CPU only
• Licensing: BSD-3-Clause (SphereSfM & COLMAP); bundled library notices in docs/SPHERESFM-THIRD-PARTY-NOTICES.txt.