sat3dgen

name	sat3dgen
version	1.0.0
description	Generate 3D scenes from satellite imagery using Sat3DGen (ICLR 2026). Sends a satellite image to the HuggingFace Space API and returns a textured 3D mesh (.obj) plus an interactive Three.js viewer.
type	script
scripts	[{"name":"generate.py","description":"Generate a 3D mesh and interactive viewer from a satellite image"}]
runtime	python3
timeout_sec	300
dependencies	["gradio_client>=1.0"]
permissions	["net"]
trigger	User provides a satellite image and wants to generate a 3D street-level scene, 3D mesh, or 3D reconstruction from it.
env_from_settings	[]

Sat3DGen — Satellite Image to 3D Scene

Generates a 3D street-level scene from a single satellite/aerial image using the Sat3DGen model (ICLR 2026).

How it works

1. Accepts a satellite image path as input (PNG/JPG, zoom level ~20 recommended)
2. Connects to the live HuggingFace Space qian43/Sat3DGen via Gradio Client
3. Submits the image for 3D mesh generation (Marching Cubes at configurable resolution)
4. Downloads the resulting .obj mesh
5. Generates an interactive Three.js HTML viewer for the mesh

Usage

skill_exec sat3dgen generate.py <image_path> [--resolution 256]

Arguments

• image_path — path to the satellite image (PNG or JPG)
• --resolution — mesh voxel resolution: 128 or 256 (default: 256)

Outputs (written to skill state directory)

• <name>_mesh.obj — textured 3D mesh
• <name>_viewer.html — interactive Three.js viewer

Requirements

• gradio_client (installed automatically via dependencies)
• Internet access to reach HuggingFace Spaces (net permission)