Academic figure practice: LaTeX methods for 5 high-frequency paper chart types

Introduction

You can polish prose for weeks, but a rough figure can hurt how reviewers read the whole paper. IEEE, ACM, Springer, and similar publishers expect vector output, consistent typography, and sensible colour use. LaTeX + TikZ meets those requirements—but many researchers avoid it because of syntax complexity. This article covers five of the most common figure types in academic papers: core TikZ patterns for each, plus how DrawFig compresses “describe → figure → code” into minutes.

---

1 — Algorithm flowcharts

Algorithm flowcharts are standard in CS papers. TikZ’s shapes.geometric library provides diamonds (decisions), rounded rectangles (start/stop), and parallelograms (I/O).

\usetikzlibrary{shapes.geometric, arrows.meta, positioning}

\begin{tikzpicture}[
  startstop/.style={rectangle, rounded corners, minimum width=2.5cm, minimum height=0.8cm, text centered, draw},
  process/.style={rectangle, minimum width=2.5cm, minimum height=0.8cm, text centered, draw},
  decision/.style={diamond, minimum width=2cm, minimum height=0.8cm, text centered, draw, aspect=2},
  arrow/.style={thick, ->, >=Stealth}
]
  \node (start) [startstop] {Start};
  \node (init) [process, below=of start] {Init parameters};
  \node (check) [decision, below=of init] {Condition met?};
  \node (update) [process, below=of check] {Update state};
  \node (stop) [startstop, below=of update] {Stop};

  \draw[arrow] (start) -- (init);
  \draw[arrow] (init) -- (check);
  \draw[arrow] (check) -- node[right] {Yes} (update);
  \draw[arrow] (check.west) -- ++(-1.5,0) |- node[above, near start] {No} (init.west);
  \draw[arrow] (update) -- (stop);
\end{tikzpicture}

Define each shape once with style, then reference it on nodes—cleaner and easier to maintain.

---

2 — Experiment comparison charts (bar / line)

Result charts often use pgfplots, a TikZ-based data-visualisation package:

\usepackage{pgfplots}
\pgfplotsset{compat=1.18}

\begin{tikzpicture}
  \begin{axis}[
    ybar,
    bar width=12pt,
    xlabel={Dataset},
    ylabel={Accuracy (\%)},
    symbolic x coords={Cora, Citeseer, Pubmed, ogbn-arxiv},
    xtick=data,
    legend style={at={(0.5,-0.2)}, anchor=north, legend columns=-1},
    ymin=60, ymax=100,
    nodes near coords,
    nodes near coords align={vertical},
  ]
    \addplot coordinates {(Cora,92.3) (Citeseer,88.7) (Pubmed,90.1) (ogbn-arxiv,85.4)};
    \addplot coordinates {(Cora,85.1) (Citeseer,82.3) (Pubmed,84.6) (ogbn-arxiv,79.2)};
    \legend{Ours, Baseline}
  \end{axis}
\end{tikzpicture}

ybar gives vertical bars; symbolic x coords uses string x-labels; nodes near coords prints values on bars—reviewers appreciate explicit numbers.

---

3 — Network topology diagrams

Graph and network papers almost always need topology figures. TikZ graphdrawing can auto-layout (LuaLaTeX). Without switching compilers, manual coordinates are the safest bet:

\begin{tikzpicture}[
  vertex/.style={circle, draw, minimum size=0.7cm, inner sep=0pt, font=\small},
  edge/.style={thick},
  node distance=2cm
]
  \node[vertex, fill=blue!20] (v1) at (0,2)   {$v_1$};
  \node[vertex, fill=blue!20] (v2) at (2,2)   {$v_2$};
  \node[vertex, fill=red!20]  (v3) at (1,0.5) {$v_3$};
  \node[vertex]               (v4) at (-1,0)  {$v_4$};
  \node[vertex]               (v5) at (3,0)   {$v_5$};

  \draw[edge] (v1) -- (v2);
  \draw[edge] (v2) -- (v3);
  \draw[edge] (v1) -- (v4);
  \draw[edge] (v3) -- (v5);
  \draw[edge, dashed] (v4) -- (v5);
  \draw[edge, bend left=25] (v1) to (v3);
\end{tikzpicture}

fill=blue!20 / fill=red!20 colour node groups; dashed marks weak ties; bend left=25 avoids overlapping straight edges.

---

4 — System architecture diagrams

Systems papers need layered modules and call relationships. TikZ fit and backgrounds draw grouped boxes:

\usetikzlibrary{fit, backgrounds, positioning}

\begin{tikzpicture}[
  module/.style={rectangle, draw, rounded corners, minimum width=2.2cm, minimum height=0.8cm, text centered, fill=white},
  arrow/.style={->, >=Stealth, thick}
]
  \node[module, fill=green!15] (ui)     {Frontend};
  \node[module, fill=yellow!15, below=of ui] (api)    {API gateway};
  \node[module, fill=blue!15, below left=1cm and 0.5cm of api] (auth)   {Auth service};
  \node[module, fill=blue!15, below right=1cm and 0.5cm of api] (biz)   {Business logic};
  \node[module, fill=orange!15, below=1.5cm of api] (db)     {Database};

  \draw[arrow] (ui) -- (api);
  \draw[arrow] (api) -- (auth);
  \draw[arrow] (api) -- (biz);
  \draw[arrow] (auth) -- (db);
  \draw[arrow] (biz) -- (db);

  \begin{scope}[on background layer]
    \node[fit=(auth)(biz), draw, dashed, rounded corners, fill=gray!5, inner sep=10pt, label=above:Microservices] {};
  \end{scope}
\end{tikzpicture}

fit computes bounding boxes; backgrounds keeps group frames behind foreground content.

---

5 — Data-flow / pipeline diagrams

ML and data-mining papers often show processing pipelines. TikZ chains strings steps quickly:

\usetikzlibrary{chains, arrows.meta}

\begin{tikzpicture}[
  box/.style={rectangle, draw, rounded corners, minimum width=1.8cm, minimum height=0.7cm, text centered, fill=#1},
  box/.default={white},
  arrow/.style={->, >=Stealth, thick}
]
  \node[box=green!15]  (raw)   {Raw data};
  \node[box=yellow!15, right=1cm of raw]  (clean) {Cleaning};
  \node[box=orange!15, right=1cm of clean](feat)  {Features};
  \node[box=blue!15,   right=1cm of feat] (train) {Training};
  \node[box=red!15,    right=1cm of train](eval)  {Evaluation};

  \draw[arrow] (raw) -- (clean);
  \draw[arrow] (clean) -- (feat);
  \draw[arrow] (feat) -- (train);
  \draw[arrow] (train) -- (eval);
\end{tikzpicture}

Each stage gets a distinct colour; arrows show data flow with high information density and low reading cost.

---

Generate all of the above with DrawFig

Every type above has workable TikZ—but in practice, coordinates and styling take many iterations. DrawFig shortens the path: Steps:

1. Open drawfig.com/editor.html and describe the figure in the AI panel—for example: “Draw a five-module system architecture: frontend → API gateway → auth and business services → shared database.”
2. AI renders to canvas; drag nodes, edit labels, adjust colours.
3. Click Export TikZ for compilable LaTeX.
4. Paste into \begin{figure}...\end{figure} and compile.

From description to TikZ code often takes under three minutes—versus half an hour or more by hand when deadlines loom.

---

Closing

Publication-quality figures are not optional—they are baseline expectations. LaTeX + TikZ is format-perfect; hand-coding is the bottleneck. Learn these five patterns, then pair them with DrawFig’s AI + canvas + TikZ export to shrink the idea → figure → paper loop. Try DrawFig for academic figures → drawfig.com/editor.html Canvas drag-and-drop editing and SVG/PNG/PDF export are free with no sign-in. TikZ import (5 credits/use), TikZ export (3 credits/use), and AI canvas generation (5 credits/use) require sign-in. You receive 30 credits daily (accumulated). See credit rules.

---

See also: TikZ graph-theory tutorial · AI dialog graph generation guide