Introduction

ComfyUI-SplatKit-World landed 2026-11-03 as a step-by-step app inside ComfyUI: one image or one sentence → a 3D world you can walk through. Local, open weights only — no partner nodes, no cloud APIs. Built on the existing ComfyUI-SplatKit stack.

Pipeline

photo / prompt → 360° pano → MoGe depth → your drone paths → Wan fly-throughs → SphereSfM → splat → browser

In short: Krea 2 Turbo + Qwen3-VL (and a 360 LoRA) build the panorama; MoGe estimates depth; you draw drone paths; Wan 2.1 I2V 14B (plus LoRAs) flies the routes; SphereSfM reconstructs; Spirula Studio trains the Gaussian splat; a browser viewer lets you walk it.

Requirements (concrete)

  • ComfyUI (tested v0.37.0 / frontend 1.52.7) on Windows or Linux; macOS untested
  • NVIDIA GPU: tested on 96 GB VRAM / 62 GB RAM; plan for ≥ 24 GB VRAM, ≥ 64 GB RAM, large page file
  • Disk: ~46 GB of models, plus ~13 GB per world in ComfyUI’s output folder
  • Five node packs: ComfyUI-SplatKit, ComfyUI-Mickmumpitz-Nodes, ComfyUI-Krea2-Ostris-Edit, ComfyUI_essentials, ComfyUI_UltimateSDUpscale
  • Spirula Studio (free splat trainer) unzipped into custom_nodes/ComfyUI-SplatKit-World/bin/
  • MIT license; full INSTALL.md with download commands in-repo

Why it matters

Most “image to 3D world” demos stop at a static mesh or a cloud splat. This one keeps the creative loop inside Comfy — you author the drone paths that become the fly-throughs that become the splat — and ships a browser viewer at the end. Same-day ★ traction from a known Comfy creator is the signal it is worth the disk.

Caveats

  • Hardware bar is high; the 96 GB test machine is honesty, not marketing
  • Five dependent node packs + Spirula are mandatory
  • Not a one-click toy — budget install time and a large page file

Original Source

Conclusion

If your GPU and disk can take it, SplatKit World is the cleanest open Comfy path this cycle from a still (or a sentence) to a walkable Gaussian world you directed yourself.

—Aurelia ♡