Prompt Details
Model
Chat - (gpt-5.5)
Token size
568
Example input
[DIAGRAM_TYPE]: Neural network architecture diagram
[DESCRIPTION]: A convolutional neural network for image classification: input image (224x224x3), two Conv+ReLU+MaxPool blocks, a Flatten layer, two Fully Connected layers, and a Softmax output with 10 classes. Show the layer name and tensor dimensions under each block.
[STYLE_PREFERENCES]: Modern flat design, soft blue and teal palette, rounded rectangles, thin gray connecting arrows, sans-serif labels
[ORIENTATION]: Left-to-right horizontal flow
Example output
## 1. CODE
```latex
\documentclass[tikz,border=10pt]{standalone}
% Standard TikZ libraries used for relative positioning and modern arrows
\usepackage{tikz}
\usetikzlibrary{positioning,arrows.meta}
% Soft blue/teal palette
\definecolor{InputBlue}{HTML}{D8ECFF}
\definecolor{ConvBlue}{HTML}{BFE3FF}
\definecolor{PoolTeal}{HTML}{BEEDE3}
\definecolor{FlatGray}{HTML}{EEF2F4}
\definecolor{FCBlue}{HTML}{D5E6FF}
\definecolor{OutputTeal}{HTML}{A9DED6}
\definecolor{EdgeGray}{HTML}{8A939B}
\definecolor{TextDark}{HTML}{24323A}
\begin{document}
\begin{tikzpicture}[
font=\sffamily,
node distance=1.25cm,
]
% Reusable styles
\tikzset{
layer/.style={
draw=EdgeGray,
line width=0.7pt,
rounded corners=6pt,
minimum width=2.25cm,
minimum height=1.15cm,
align=center,
text=TextDark,
inner sep=6pt
},
input/.style={layer, fill=InputBlue},
conv/.style={layer, fill=ConvBlue},
pool/.style={layer, fill=PoolTeal},
flatten/.style={layer, fill=FlatGray},
fc/.style={layer, fill=FCBlue},
output/.style={layer, fill=OutputTeal},
arrow/.style={
-{Stealth[length=2.4mm,width=1.8mm]},
draw=EdgeGray,
line width=0.75pt,
rounded corners=2pt
},
layerlabel/.style={
align=center,
text=TextDark,
font=\sffamily\footnotesize,
inner sep=2pt
}
}
% nodes
\node[input] (input) {Input\\Image};
\node[conv, right=of input] (conv1) {Conv\\ReLU};
\node[pool, right=of conv1] (pool1) {MaxPool};
\node[conv, right=of pool1] (conv2) {Conv\\ReLU};
\node[pool, right=of conv2] (pool2) {MaxPool};
\node[flatten, right=of pool2] (flatten) {Flatten};
\node[fc, right=of flatten] (fc1) {Fully\\Connected};
\node[fc, right=of fc1] (fc2) {Fully\\Connected};
\node[output, right=of fc2] (softmax) {Softmax\\Output};
% labels
\node[layerlabel, below=0.22cm of input] {224$\times$224$\times$3};
\node[layerlabel, below=0.22cm of conv1] {Conv Block 1\\112$\times$112$\times$32};
\node[layerlabel, below=0.22cm of pool1] {56$\times$56$\times$32};
\node[layerlabel, below=0.22cm of conv2] {Conv Block 2\\28$\times$28$\times$64};
\node[layerlabel, below=0.22cm of pool2] {14$\times$14$\times$64};
\node[layerlabel, below=0.22cm of flatten] {12,544};
\node[layerlabel, below=0.22cm of fc1] {512};
\node[layerlabel, below=0.22cm of fc2] {128};
\node[layerlabel, below=0.22cm of softmax] {10 classes};
% edges
\draw[arrow] (input) -- (conv1);
\draw[arrow] (conv1) -- (pool1);
\draw[arrow] (pool1) -- (conv2);
\draw[arrow] (conv2) -- (pool2);
\draw[arrow] (pool2) -- (flatten);
\draw[arrow] (flatten) -- (fc1);
\draw[arrow] (fc1) -- (fc2);
\draw[arrow] (fc2) -- (softmax);
\end{tikzpicture}
\end{document}
```
## 2. COMPILE NOTES
Assumption: the convolution blocks use stride/pooling choices that reduce the spatial tensor sizes as shown, with channel counts 32 then 64.
Compile with **pdfLaTeX**, **XeLaTeX**, or **LuaLaTeX**. No external images or nonstandard TikZ libraries are required. A standard TeX Live or MiKTeX installation with TikZ/PGF is sufficient.
## 3. CUSTOMIZE
* Change the palette in the `\definecolor` lines near the top, especially `ConvBlue`, `PoolTeal`, `FCBlue`, and `OutputTeal`.
* Adjust spacing by changing `node distance=1.25cm` in the `tikzpicture` options.
* Edit block size and corner radius in the `layer/.style` key: `minimum width`, `minimum height`, and `rounded corners`.
* Change labels and tensor dimensions in the `% labels` section under each node.
* To switch orientation, replace `right=of ...` with `below=of ...` in the `% nodes` section and redraw the flow vertically.
Or get it with Select10 Select prompts/mo — under $2 each
Learn more
Instant access
Commercial use
Money‑backBy purchasing this prompt, you agree to our terms of service
3
Views
GPT-5.5
Turn plain-English descriptions into clean, compile-ready LaTeX TikZ code. Built for researchers and academics who need publication-quality diagrams without wrestling with syntax.
🧠 Neural network & ML architecture diagrams
🔀 Workflows, flowcharts & pipelines
🌳 Decision trees & hierarchies
📐 Self-contained, error-free .tex output
🎨 Customizable style, layout & orientation
⚡ Works in Overleaf, pdfLaTeX, XeLaTeX, LuaLaTeX
...more
Added 1 week ago
