Multi-Layer Perceptron (MLP)
Configurable input, hidden, and output neurons with activation labels and residual skip connections.
Design publication-ready deep learning architectures with visual node and layer controls: MLPs with skip connections, 3D CNN feature map cuboids, Multi-Head Attention blocks, and LSTM cells.
IhateLaTeX runs entirely in your browser via a WebAssembly pdfTeX engine — no servers, no sign-up, no uploads. Every tool below is 100% free with unlimited use. The Neural Network Generator produces clean, idiomatic LaTeX TikZ code with named styles and layer alignments for your machine learning papers.
Configurable input, hidden, and output neurons with activation labels and residual skip connections.
Isometric volumetric feature cuboids showing Conv2D, MaxPool, and dense layer projections.
Multi-Head Attention, Scaled Dot-Product, Positional Encodings, and Add & LayerNorm stacks.
Contracting and expansive convolutional pathways with horizontal skip concatenation.
IEEE Grayscale, Modern AI Indigo, Brand Blue, and High-Contrast palettes.
Copy pure TikZ snippets, standalone compilable documents, lossless SVGs, or 300 DPI PNGs.
Select the MLP or Transformer architecture in the studio, customize neuron counts and layer labels, and click Copy TikZ Code to paste directly into Overleaf or TeXstudio.
Yes. The 3D CNN generator produces isometric feature cuboids with dimension tags (e.g. 224x224x3, 112x112x64) and receptive field projection lines.
Yes, 100% free and client-side. Unpublished model architectures stay strictly on your local machine.