01

Paint the brain

Paint neurons. Release to stimulate.

Ready to load
L
RestingStimulatedFiring
R
Actions
02

Watch the response

At rest
Drosophila melanogaster1 mm
Walking drive0 Hz
Turning0°/s
Escape drive0 Hz

Neural Canvas

Load the neural weights and prepare the scene to start painting.

166,700 neurons · 76 MB neural weights

Runs locally in your browser. Downloaded weights are cached for next time.

Experiment settings

Model & sources

166,700 neurons from MaleCNS v1.0. Dataset IDs identify individual cells; they are not consecutive neuron numbers.

Recorded connectivity and adapted spiking dynamics drive crafted walking, turning, and flight animations. The movements are illustrative, not validated predictions of fly behavior. The scanned body is a female NeuroMechFly specimen.

Turn actions stimulate LC9 and DNa02 on the selected side. Directly stimulated neurons can still fire when recurrent transmission is disabled.

About this tool

Mapping every neuron and every connection in an animal's nervous system — a full "connectome" — was, for decades, too laborious to attempt for anything more complex than a worm with a few hundred neurons. Advances in high-resolution electron microscopy and automated image analysis changed that: researchers have now reconstructed the complete wiring diagram of a fruit fly's brain and nerve cord, identifying roughly 166,700 individual neurons and well over a hundred million synaptic connections between them — arguably the most detailed map of an animal's central nervous system produced to date.

This page lets you explore that map directly: paint a patch of neurons on the brain and watch the stimulus ripple outward along real recorded connections, then see how that activity translates into movement on an anatomically accurate, three-dimensional fly body — walking, turning, or taking off, depending on which circuits were engaged. It turns an abstract wiring diagram into something you can prod, watch respond, and get an intuitive feel for, in a way a static diagram or a research paper never could.

Everything — loading the connectome, simulating neural activity, and animating the fly — runs locally in the browser once the data has been downloaded once and cached; nothing about the session is uploaded anywhere. The simulation prefers WebGPU for speed and falls back to a slower CPU-only mode automatically on devices without it.

Note: this specific tool's own on-screen controls are in English only, since it is an independent, unmodified research demo embedded here rather than a page authored for this site.

Credits & Licenses

This tool is an unmodified copy of the “Neural Canvas” demo by Xenova, embedded here as downloaded; its application code is MIT licensed. Neuron and connectivity data are from the MaleCNS v1.0 dataset (CC BY 4.0), credited to FlyEM / HHMI Janelia, the University of Cambridge, the MRC Laboratory of Molecular Biology, and Google Research. The spiking neural model is adapted from Shiu et al., 2024; the animated fly body follows the NeuroMechFly project (EPFL); and on-device inference uses Hugging Face’s WebGPU kernels.