Neural Network Diagram Generator for MLP, CNN & Transformers

Neurons, layers, weights, and connections

• Neurons are the nodes of the network — each one takes inputs, combines them, and passes the result on. They are drawn as circles.
• Layers group neurons into columns. The input layer receives your data, one or more hidden layers transform it, and the output layer produces the prediction.
• Connections are the lines between neurons. In a fully-connected (dense) network, every neuron in one layer links to every neuron in the next.
• Weights are the strengths of those connections — the numbers the network learns during training. The diagram shows the wiring; training sets the weights.

Two ways to make a neural network diagram here

• Precise mode: type your layer sizes (for example "4, 6, 6, 2", or one layer per line as "Input, 4") and the tool draws an exact, evenly-spaced, fully-connected feedforward network with every layer labeled — accurate every time, no dragging.
• AI illustration mode: describe a more complex architecture — a CNN, Transformer, RNN, GAN, or autoencoder — in plain English and the tool generates a polished, presentation-ready diagram with labeled blocks and data-flow arrows.
• Use precise mode for standard MLPs where the layer structure must be exactly right; use AI mode when you need the specialized building blocks of deep-learning architectures.

Feedforward vs CNN, RNN, and Transformer

A feedforward network (a multilayer perceptron, or MLP) passes data straight through stacked dense layers and is the right model for tabular data and simple classification — this is exactly what the precise mode renders. Convolutional neural networks (CNNs) add convolution and pooling layers for images. Recurrent networks (RNNs and LSTMs) loop information across time steps for sequences and text. Transformers replace recurrence with self-attention and power most modern large language models. These architectures have specialized blocks that go beyond plain layers of circles, so the AI illustration mode is the better fit for drawing them.

How to make a neural network diagram

• Decide whether your network is a standard feedforward model or a specialized architecture (CNN, RNN, Transformer).
• For a feedforward net, open the precise mode and enter your layer sizes — a comma list, or one named layer per line.
• The tool evenly spaces the layers, draws every neuron-to-neuron connection, and labels each layer with its name and neuron count.
• For a CNN, RNN, or Transformer, switch to AI illustration mode and describe the architecture, including the layers and data flow you need.
• Export a clean SVG to drop into slides, reports, papers, or documentation.

When to use the AI illustration mode

Reach for AI illustration mode when your architecture needs more than plain layers of neurons — convolutional and pooling stages, attention mechanisms, gated recurrent cells, an encoder–decoder split, or a labeled training loop. The precise mode is purpose-built for standard feedforward / MLP networks where exact, fully-connected geometry matters; the AI mode handles the visual complexity of CNNs, Transformers, RNNs, GANs, and autoencoders that a circles-and-lines layout cannot capture on its own.