A lightweight WebGPU-powered visualizer for experimenting with neural cellular automata. This project uses a GPU compute shader to apply 5×5 convolution filters over a grid of pixels, allowing real-time updates to the weights and activation functions that control the behavior of the automata.
Hosted on GitHub Pages here
- Each pixel is updated using a 5×5 convolution per color channel (R, G, B).
- You can modify 225 total weights (3 output × 3 input × 5 pixels × 5 pixels) in real time.
- Supports a nearly unlimited variety of activation functions, if you can write it in WGSL you can use it!
- Built with WebGPU and React.
- Node.js ≥ 18
- A browser and device with WebGPU support. WebGPU is not enabled by default on some browsers
git clone https://github.com/Stermere/Neural-Automata-Playground/
cd Neural-Automata-Playground
npm install# Add -- --host to host it on your local network
npm run dev


