Neural Network Diagram Generator

Figran

Choose an architecture, set the layers, and download a diagram for your paper.

  • Four architecture types
  • Live preview
  • SVG and PNG
Architecture
View
Layers
Direction
Edit in canvas

Neural network architecture diagram examples

Load a familiar architecture, then adjust it for your model.

Perceptron diagram

MLP diagram

Autoencoder architecture diagram

CNN architecture diagram (LeNet-5)

AlexNet architecture diagram

VGG16 architecture diagram

ResNet architecture diagram (residual block)

LSTM diagram

RNN architecture diagram

Transformer architecture diagram

How to draw a neural network diagram

  1. 1Choose an architecture

    Start with a fully connected network, CNN, recurrent network, or Transformer.

  2. 2Set layers and connections

    Add layer names, neuron counts or feature map dimensions. Label the operations that connect them.

  3. 3Download your diagram

    Use SVG for an editable vector figure or PNG when your manuscript needs a raster image.

Where the model architecture diagram goes in your paper

Place a deep learning architecture diagram near the Methods text that explains the model.

Illustrative manuscript page

2. Methods

We use a feed forward model with two hidden layers. Figure 1 summarizes the input, hidden, and output layers used for prediction.

Figure 1. Overview of the proposed model.

Neural network diagram FAQ

What is a neural network diagram?

A neural network diagram, or neural net diagram, shows the layers of an artificial neural network and the connections between them. A simple feed forward network has input, hidden, and output layers; other architectures use feature maps, recurrent cells, or attention blocks.

How do I make a CNN architecture diagram?

A convolutional neural network (CNN) architecture diagram lists the layers in order. Choose CNN, enter the layers and feature map dimensions in order, and label the operations between them. Select 3D feature maps or Flat layer blocks.

Can I draw an LSTM or RNN diagram?

Choose LSTM / RNN. Cell shows the gates and states inside an LSTM cell. Unrolled supports 3–6 time steps, 1–3 stacked layers, RNN, LSTM or GRU cells, unidirectional or bidirectional recurrence, and many-to-many or many-to-one outputs.

Is this a neural network visualizer?

This tool draws architecture schematics from the layers you specify. It does not visualize trained weights or activations. To inspect a supported model file, use Netron.

How is this different from NN-SVG or PlotNeuralNet?

NN-SVG provides parameterized network drawings in the browser. PlotNeuralNet uses Python and LaTeX. This tool provides forms for four architecture types, SVG and PNG downloads, and an editable canvas without writing code.

Can I edit the diagram in PowerPoint, Word or Illustrator?

SVG keeps the diagram as vector shapes and text. Import the SVG into PowerPoint, Word, or Illustrator to edit it. PNG provides a raster alternative.

Is it free?

SVG and 300 DPI PNG downloads are free and do not require sign-in. Editing in the canvas requires a free account.

Ready to draw your model architecture?

Choose an example, set the layers, and download a diagram for your manuscript.

Create your diagram