Skip to main content

LaTeX Posters and Conference Materials: Large Format Printing

Created: March 7, 2026 CalmOps 10 min read
Table of Contents

Introduction

Conference posters demand large-format printing with professional typography, crisp vector graphics, and precise layout control. LaTeX handles all of this natively — text does not blur when scaled, equations render correctly at any size, and your bibliography is automatically formatted.

This guide covers poster creation from scratch: choosing paper dimensions, using the major poster classes, embedding TikZ diagrams and images, managing references, and applying sound design principles. Every example is compilable as-is on a modern TeX distribution (TeX Live 2024+ or MiKTeX).

Choosing Paper Size

Academic conferences almost always specify a poster size. The two most common formats are A0 (33.1 × 46.8 in / 841 × 1189 mm) and A1 (23.4 × 33.1 in / 594 × 841 mm). Some US-based conferences request 36×48 in or 42×42 in.

% A0 portrait (standard for most European conferences)
\documentclass[a0paper,portrait]{a0poster}

% A1 landscape (common for smaller poster sessions)
\documentclass[a1paper,landscape]{a0poster}

% Custom US size (36 in × 48 in)
\documentclass[paperwidth=36in,paperheight=48in,landscape]{baposter}

A0 gives you roughly eighteen A4 pages of content area — far more than a slide deck. Do not fill every square inch. Leave generous margins (3–5 cm) and whitespace between sections.

Portrait vs Landscape

Portrait posters read naturally top-to-bottom and work well for narrow, deep content. Landscape posters let you arrange three or four columns side-by-side, which suits sprawling experimental workflows (introduction → methods → results → conclusion). Most A0 posters use portrait; most US-format posters use landscape.

a0poster Class: Full Control

The a0poster class is the oldest and most widely used LaTeX poster class. It behaves like a standard article but scales fonts and paper appropriately for large format.

\documentclass[a0paper,portrait]{a0poster}
\usepackage[utf8]{inputenc}
\usepackage{multicol}
\usepackage{graphicx}
\usepackage{amsmath,amssymb}
\usepackage{tikz}
\usepackage[margin=3cm]{geometry}

\title{Deep Learning Approaches for Protein Structure Prediction}
\author{Alice Researcher\textsuperscript{1}, Bob Scientist\textsuperscript{2}}
\institute{\textsuperscript{1}University of Science, \textsuperscript{2}Bioinformatics Institute}

\begin{document}

\maketitle

\begin{multicols}{3}

\section{Introduction}
Proteins fold into three-dimensional structures that determine their function.
Predicting these structures from amino-acid sequences remains a central
challenge in computational biology. Recent advances in deep learning have
dramatically improved prediction accuracy.

\section{Methods}
We trained a transformer-based model on 50,000 known protein structures.
The architecture uses multi-head attention to capture long-range residue
interactions. Training was performed on 8 NVIDIA A100 GPUs for 72 hours.

\section{Results}
\begin{figure}[h]
\centering
\includegraphics[width=\linewidth]{results-plot.pdf}
\caption{TM-score distribution across test set proteins.}
\end{figure}

Our model achieved a median TM-score of 0.89, outperforming the previous
state-of-the-art by 12\%.

\end{multicols}

\end{document}

Font Sizing

The a0poster class provides scaled font commands: \huge maps to approximately 50 pt, \Huge to 60 pt. For finer control, use \fontsize{size}{leading}\selectfont:

\fontsize{48}{60}\selectfont Section Title

Recommended body text size is 28–36 pt. Section headings should be 40–48 pt. The poster title should be 72–96 pt.

Including Figures

Scale images using \linewidth or \columnwidth rather than absolute dimensions — this keeps proportions correct when you switch paper sizes.

\begin{figure}[h]
\centering
\includegraphics[width=0.9\columnwidth]{architecture-diagram.pdf}
\caption{Model architecture showing encoder-decoder with attention layers.}
\label{fig:architecture}
\end{figure}

Always convert raster images to at least 300 DPI at final print size. A 10 cm wide image at 300 DPI needs to be 1181 pixels wide. Vector graphics (PDF, EPS) are preferred because they scale infinitely.

baposter Class: Designer-Friendly

The baposter class provides color-coded header boxes, rounded corners, drop shadows, and column-aware layout. It requires a poster environment and uses headerbox commands for each section.

\documentclass[paperwidth=36in,paperheight=48in,landscape]{baposter}
\usepackage{graphicx}
\usepackage{amsmath}
\usepackage[latin9]{inputenc}
\usepackage[T1]{fontenc}

\definecolor{boxblue}{HTML}{1E3A5F}
\definecolor{boxgold}{HTML}{C9A84C}

\begin{poster}{
    columns=4,
    colspacing=1em,
    headerheight=0.1\textheight,
    background=user,
    bgColorOne=white,
    borderColor=boxblue,
    headerColorOne=boxblue,
    headerFontColor=white,
    boxColorOne=white,
    textborder=rounded,
    eye catching=true,
    eyecatcher=true
}
{\includegraphics[height=6em]{logo.pdf}}
{\Huge Quantum Error Correction with Surface Codes}
{Alice Researcher, Bob Scientist}
{University of Science}

\headerbox{Abstract}{name=abstract,column=0,row=0}{
Quantum error correction is essential for fault-tolerant computing.
Surface codes offer high thresholds and planar geometry suitable for
superconducting qubit architectures.
}

\headerbox{Methods}{name=methods,column=0,row=1,span=1}{
We implemented a minimum-weight perfect matching decoder running on
a classical co-processor with microsecond latency.
}

\headerbox{Experimental Setup}{name=setup,column=1,row=0,span=2}{
\begin{figure}[h]
\centering
\includegraphics[width=0.8\linewidth]{chip-photo.jpg}
\caption{17-qubit superconducting processor.}
\end{figure}
}

\headerbox{Results}{name=results,column=1,row=1,span=2}{
\begin{align*}
p_{\text{threshold}} &= 0.57\% \\
d &= 3,\;5,\;7,\;9 \\
\mathcal{L} &\propto d^{2}
\end{align*}
Logical error rate decreases exponentially with code distance.
}

\headerbox{Conclusion}{name=conclusion,column=3,row=0,span=1}{
Surface codes are compatible with existing superconducting hardware.
Future work will explore flag qubit schemes for improved thresholds.
}

\headerbox{References}{name=refs,column=3,row=1,span=1}{
\begin{small}
[1] Kitaev, A. (2003). Fault-tolerant quantum computation by anyons.
\emph{Annals of Physics}, 303(1), 2–30.

[2] Fowler, A. G. et al. (2012). Surface codes: Towards practical
large-scale quantum computation. \emph{Phys. Rev. A}, 86(3), 032324.
\end{small}
}

\end{poster}

The columns parameter in \begin{poster} sets the grid count. Each headerbox occupies a rectangular cell defined by column, row, and span. This grid system gives you a magazine-style layout far beyond what a0poster can achieve.

Color Customization

Define colors with \definecolor and reference them in the poster options. Keep contrast high — dark headers with white text, light backgrounds with dark body text. Avoid red-green combinations since 8% of male readers have color vision deficiency.

\definecolor{headerbg}{HTML}{2C3E50}
\definecolor{accent}{HTML}{E74C3C}
\definecolor{bodybg}{HTML}{ECF0F1}

beamerposter Class

The beamerposter class extends Beamer to poster dimensions. Use it when you already have a Beamer presentation and want a consistent look between slides and poster.

\documentclass[final,hyperref={pdfpagelabels=false}]{beamer}
\usepackage[orientation=portrait,size=a0,scale=1.4]{beamerposter}

\usetheme{default}

\title{Poster Title}
\author{Author}
\institute{Institute}

\begin{document}
\begin{frame}{}
\begin{columns}[T]
\begin{column}{0.32\textwidth}
\block{Introduction}{Content...}
\end{column}
\begin{column}{0.32\textwidth}
\block{Methods}{Content...}
\end{column}
\begin{column}{0.32\textwidth}
\block{Results}{Content...}
\end{column}
\end{columns}
\end{frame}
\end{document}

The scale parameter controls font magnification. A scale of 1.4 is typical for A0. Increase to 1.8 if you are targeting A1.

tikzposter Class

The tikzposter class builds every poster element from TikZ, giving you complete control over borders, shadows, and decorations. It ships with several built-in themes.

\documentclass[25pt, a0paper, portrait, margin=2cm]{tikzposter}
\usetheme{Desert}

\title{TikZ Poster Example}
\author{Author}
\institute{Institute}

\begin{document}
\maketitle

\block{Section Title}{
\begin{tikzfigure}[Attention Mechanism]
\begin{tikzpicture}
\node[draw,circle,fill=blue!20] (q) at (0,0) {Query};
\node[draw,circle,fill=red!20]  (k) at (2,0) {Key};
\node[draw,circle,fill=green!20] (v) at (4,0) {Value};
\draw[->,thick] (q) -- (k);
\draw[->,thick] (k) -- (v);
\draw[->,thick,dashed] (q) -- (v);
\end{tikzpicture}
\end{tikzfigure}
}

\block{Results}{
\begin{tikzfigure}[Benchmark Comparison]
...
\end{tikzfigure}
}
\end{document}

Comparison Table: Poster Classes

% This table belongs in your project README or documentation
Feature a0poster baposter beamerposter tikzposter
Standard paper sizes A0, A1, A2 Custom (inch) A0, A1, custom A0, A1, A2, custom
Multi-column layout multicols Built-in grid Beamer columns block-based
Header boxes No Yes, with color \block macro \block macro
Built-in themes No 4 variants Beamer themes 6+ themes
TikZ integration Manual Manual Manual Native
Bibliography support biblatex/bibtex Manual only biblatex/bibtex biblatex/bibtex
Learning curve Low Medium Medium (needs Beamer) Medium
Best for Simple layouts Complex grids Beamer consistency Heavy TikZ usage

Adding TikZ Diagrams

TikZ produces vector diagrams that scale perfectly to poster dimensions. Wrap TikZ code in a figure environment or use tikzposter’s tikzfigure.

\begin{figure}[h]
\centering
\begin{tikzpicture}[scale=1.5, every node/.style={font=\large}]
\draw[thick,->] (0,0) -- (6,0) node[right] {Epochs};
\draw[thick,->] (0,0) -- (0,4) node[above] {Loss};
\draw[blue,thick] plot coordinates {
    (0,3.5) (1,2.8) (2,2.0) (3,1.2) (4,0.8) (5,0.6)
};
\draw[red,dashed,thick] plot coordinates {
    (0,3.8) (1,3.2) (2,2.5) (3,1.8) (4,1.4) (5,1.2)
};
\node[blue] at (4.5,1.0) {Training};
\node[red] at (4.5,1.8) {Validation};
\end{tikzpicture}
\caption{Training and validation loss curves.}
\label{fig:loss}
\end{figure}

For complex diagrams (neural networks, system architectures, flowcharts), draw them externally in Inkscape and export as PDF, or use the svg package to include SVG figures directly.

Displaying Algorithms and Equations

Conference posters often highlight a key algorithm or mathematical formulation. Use algorithm2e or plain align environments:

\subsection{Training Algorithm}

\begin{algorithm}[h]
\caption{Stochastic Gradient Descent}
\KwIn{Training set $\mathcal{D}$, learning rate $\eta$, epochs $T$}
\KwOut{Trained parameters $\theta$}
Initialize $\theta \leftarrow \theta_0$\;
\For{$t = 1$ to $T$}{
    Sample batch $\mathcal{B} \subset \mathcal{D}$\;
    Compute gradient $g = \frac{1}{|\mathcal{B}|} \sum_{x \in \mathcal{B}} \nabla_\theta \mathcal{L}(x; \theta)$\;
    Update $\theta \leftarrow \theta - \eta g$\;
}
\Return{$\theta$}\;
\end{algorithm}

Pair algorithms with a key equation block using align*:

\begin{align*}
\mathcal{L}_{\text{total}} &= \mathcal{L}_{\text{CE}} + \lambda \mathcal{L}_{\text{reg}} \\
\mathcal{L}_{\text{CE}} &= -\frac{1}{N} \sum_{i=1}^{N} \sum_{c=1}^{C} y_{i,c} \log p_{i,c} \\
\mathcal{L}_{\text{reg}} &= \frac{1}{2} \| \theta \|_2^2
\end{align*}

Including Tables

Tables on posters need large fonts and minimal clutter. Use booktabs for clean rules:

\begin{table}[h]
\centering
\begin{tabular}{lccc}
\toprule
Model & Accuracy (\%) & Precision (\%) & Recall (\%) \\
\midrule
CNN Baseline  & 84.2 & 83.1 & 82.9 \\
ResNet-50     & 91.7 & 91.0 & 90.8 \\
Ours          & \textbf{94.3} & \textbf{93.9} & \textbf{93.7} \\
\bottomrule
\end{tabular}
\caption{Benchmark results on the test set. Best values in bold.}
\end{table}

Bibliography Management

Use biblatex with a .bib file. Keep references minimal — 5–10 key citations per poster.

\usepackage[style=numeric,backend=biber,maxnames=3]{biblatex}
\addbibresource{references.bib}

At the bottom of your poster:

\section{References}
\printbibliography[heading=none]

For baposter, you must insert references manually in a headerbox since the class does not support \printbibliography natively.

Poster Design Principles

Font sizes: Poster title 72–96 pt, section headers 40–52 pt, body text 28–36 pt, captions 22–26 pt. Readability drops sharply below 24 pt on A0.

Contrast: Use dark text on a light background or light text on a dark background. Avoid low-contrast combinations like gray-on-white or yellow-on-white.

Whitespace: Each section block needs roughly 40% content and 60% empty space around it. Crowded posters repulse readers.

Visual hierarchy: The title should dominate. Secondary headings guide the reader. Figures and tables provide visual breaks. Three to four columns give the best scan pattern.

Color limit: Use two or three colors maximum plus black and white. A single accent color for headers and key data points is sufficient.

Comparison with PowerPoint and Canva

Factor LaTeX PowerPoint / Canva
Typography Professional, hyphenation Manual kerning, no hyphenation
Equations Native, perfect Insert as images, blurry
Bibliography Automatic (biblatex) Manual, no formatting
Vector graphics Native TikZ, PDF Imported, may rasterize
Version control Plain text, git-friendly Binary files, no diffs
Collaboration GitHub, Overleaf Shared links, proprietary
Learning curve Steep Shallow
Output consistency Identical on any system Varies by version and OS

LaTeX delivers superior typography and math, but requires upfront setup. PowerPoint or Canva work for non-technical posters with minimal equations. For research conferences, use LaTeX.

Complete, Compilable Poster Example

Below is a full A0 portrait poster combining all techniques discussed:

\documentclass[a0paper,portrait]{a0poster}
\usepackage[utf8]{inputenc}
\usepackage[T1]{fontenc}
\usepackage{multicol}
\usepackage{graphicx}
\usepackage{amsmath,amssymb}
\usepackage{tikz}
\usepackage{booktabs}
\usepackage{caption}
\usepackage[margin=3cm]{geometry}

\title{\Huge Federated Learning for Medical Image Segmentation}
\author{\Large Alice Researcher\textsuperscript{1}, Bob Scientist\textsuperscript{2}}
\institute{\large \textsuperscript{1}Department of Computer Science, Stanford University \\
           \textsuperscript{2}School of Medicine, Johns Hopkins University}

\begin{document}

\maketitle

\begin{multicols}{3}

\section{Motivation}
Medical imaging data is siloed across institutions due to privacy regulations.
Federated learning trains a shared model without transferring raw patient data.

\section{Federated Averaging (FedAvg)}
\begin{enumerate}
\item Server initializes global model $\theta^0$.
\item Round $r$: server sends $\theta^{r}$ to $K$ clients.
\item Each client trains on local data: $\theta^{r+1}_k \gets \text{LocalUpdate}(\theta^r)$.
\item Server aggregates: $\theta^{r+1} \gets \frac{1}{K} \sum_{k=1}^K \theta^{r+1}_k$.
\end{enumerate}

\section{Loss Function}
\begin{align*}
\mathcal{L}(\theta) &= \frac{1}{N} \sum_{i=1}^N \ell(f(x_i; \theta), y_i) \\
\ell(p, y) &= -\sum_{c=1}^C y_c \log p_c
\end{align*}

\section{Experimental Setup}
\begin{figure}[h]
\centering
\begin{tikzpicture}[scale=0.8]
\draw[thick,fill=blue!10] (0,0) rectangle (3,2) node[midway] {Hospital A};
\draw[thick,fill=green!10] (4,0) rectangle (7,2) node[midway] {Hospital B};
\draw[thick,fill=orange!10] (8,0) rectangle (11,2) node[midway] {Hospital C};
\draw[thick,->] (3,1) -- (4,1);
\draw[thick,->] (7,1) -- (8,1);
\draw[thick,->] (5.5,2) -- (5.5,3) -- (1.5,3) -- (1.5,2);
\node at (5.5,3.3) {Server};
\end{tikzpicture}
\caption{Federated learning topology with three hospital clients.}
\end{figure}

\section{Results}
\begin{table}[h]
\centering
\begin{tabular}{lcc}
\toprule
Method & Dice (\%) & HD95 (mm) \\
\midrule
Local only & 72.4 & 15.3 \\
FedAvg & 85.1 & 8.7 \\
FedProx & 86.3 & 7.9 \\
\bottomrule
\end{tabular}
\caption{Segmentation performance on brain tumor MRI.}
\end{table}

\begin{figure}[h]
\centering
\includegraphics[width=\linewidth]{segmentation-example.png}
\caption{Qualitative comparison of segmentation masks.}
\end{figure}

\section{Conclusion}
Federated learning achieves 86\% Dice on medical image segmentation
without centralizing patient data. Future work targets heterogeneous
client data distributions.

\end{multicols}

\end{document}

Troubleshooting

Fonts too small on A0

Pass the scale option to beamerposter or use explicit \fontsize commands. Do not rely on the default \normalsize — it is typically 11 pt, invisible on a poster.

Images too large or too small

Use width=\linewidth or width=0.9\columnwidth instead of absolute dimensions like width=10cm. This lets the image adapt when you change column count or paper size.

“Font XYZ not found”

Install the missing font package, or switch to the default Computer Modern with \usepackage[T1]{fontenc}\usepackage{lmodern}. Most conference venues render CM without issues.

Memory exhausted

Large TikZ diagrams consume TeX memory. Increase memory with pdflatex --extra-mem-top=10000000 or compile with LuaLaTeX, which has dynamic memory allocation.

baposter header text clipping

Reduce the headerheight option or decrease the header font size. The default header box leaves limited vertical space for tall descenders like “g” or “y”.

Overfull \hbox warnings

These are normal on posters because multicols cannot always break perfectly. Reduce \columnsep to 1em and increase \tolerance to 4000: \tolerance=4000.

Bibliography not printing

Ensure you run biber (not BibTeX) after pdflatex: pdflatex poster && biber poster && pdflatex poster. For baposter, skip \printbibliography and format references manually.

Resources

Comments

Share this article

Twitter Bluesky Mastodon LinkedIn Facebook Reddit

Scan to read on mobile