SPH-NCA Interactive Demo

Train Once, Generate Anywhere: Discretization Agnostic Neural Cellular Automata using SPH Method

Hyunsoo Kim, Jinah Park

Computer Graphics and Visualization Lab, KAIST

SIGGRAPH 2025 Posters

Preprint Code Poster

Weight

Inference settings

Point Pattern:

Resolution: 64

Noise: 0.0

Spatially Varying Noise

Inference control

Interaction

Brush Radius: 0.16

Brush Type:

Visualization

Color Mode:

About

See how SPH-NCA evolves over time with different weights and parameters using this interactive demo. All pretrained weights are trained only on a square grid with 64x64 resolution.

This implementation of SPH-NCA runs entirely in your browser, using JavaScript.

Example Videos

RGBA Image

Trained on a 128x128 grid, generating on a 512x512 grid

Exemplar-guided Texture

Trained on a 128x128 grid, generating on a irregular grid

Text-guided Texture

Trained on a 128x128 grid, generating on a 3D surface