demo · v147 · WebNN

Neural Network Visualizer

Drag nodes onto the canvas to build a neural network, then compile and run it with WebNN's MLGraphBuilder. Output activations are colored by magnitude. Falls back to JS simulation when WebNN is unavailable.

Origin trial: WebNN requires chrome://flags/#web-machine-learning-neural-network in Chrome 147. When the flag is off, inference is simulated in JavaScript — the graph structure and activations are identical.
Add node:
input
dense
relu
sigmoid
softmax
output
Presets:
click to place node · drag to connect · click node to delete
No graph compiled
Graph summary
Add nodes and click Compile & Run.
Output activations
Run inference to see activations.

API trace

how it works

  1. Click a node type button, then click on the canvas to place it.
  2. Drag from one node's edge to another to connect them.
  3. Click Compile & Run to call MLGraphBuilder and execute inference.
  4. Output activations are displayed as colored bars — blue = high, rose = low/negative.
  5. Node weights are randomised; re-run to see different activations.

see also

implementation reference

Need the exact API surface, compatibility boundaries, errors, lifecycle, and source links? Read the matching gendn reference ↗