Case study — how a research lab sped up figure work with DrawFig

Published: 2026-03-10 Tags: user story, research figures, graph visualisation, efficiency Reading time: ~8 min

---

Background (anonymised)

Persona: “Dr. Li” (composite) Setting: CS department, research university (China) Focus: Social graphs & graph neural networks DrawFig usage window: ~2 months Prior stack: Microsoft Visio + Python NetworkX screenshots + hand-written TikZ

---

Pain points

1 — Fragmented toolchain

• Visio: manual placement for every vertex/edge on dense graphs
• NetworkX: quick plots, but raster exports rarely meet journal expectations
• Raw TikZ: two to three hours for a single intricate topology

2 — Painful revision loops

• Supervisors request colour tweaks, relayouts, label nudges
• Each pass meant redoing Visio geometry or rewriting TikZ blocks
• File names like figure_v1_final_really_final.tex multiplied

3 — Style drift across figures

• Multiple co-authors → inconsistent fonts, node sizes, palettes
• Reviewer comment pattern: “Figures need a unified visual language.”

4 — Onboarding cost

• Every newcomer had to learn Visio and plotting scripts and TikZ basics
• Chinese-language guidance was sparse for TikZ-heavy workflows

---

How DrawFig helped — phase 1 (week 1)

Days 1–2 — Orientation - Create accounts, explore templates (flow, state machine, networks) - Practice drag-and-drop editing Days 3–4 — Bring your own graph data - Export NetworkX structures as JSON lists of nodes/edges - Import into DrawFig-style workflows and apply force-directed layout Days 5–7 — First deliverable - Recreate Figure 1 (community detection output) - Apply colour-blind–friendly presets where available - Export PDF + TikZ for direct inclusion in LaTeX Outcome - ~3 hours of drawing compressed to ~30 minutes for comparable quality - TikZ output needed only light cosmetic edits - Supervisor sign-off on first major revision

---

Phase 2 — scaling (weeks 2–4)

Team conventions - Fixed vertex geometry (e.g. 8 mm circles, 1 pt stroke) - Shared font (Times New Roman 10 pt) for print submissions - Palette locked to the “academic” preset when available - Library of reusable .drawio / project templates checked into Git Batch workflows - Parameter sweeps expressed as JSON configs (graph size, density, communities) - Parallel exports (figure_3a.pdf, figure_3b.pdf, …) Collaboration split - PI owns layout language & final QA - Student A handles imports + first-pass layout - Student B polishes labels + export settings - Git tracks source diagrams alongside LaTeX Outcome - Eight paper figures finished inside two weeks - Visual consistency across the submission package - New teammate productive in ~3 days instead of ~4–6 weeks

---

Phase 3 — integration (month 2)

LaTeX pipeline

\begin{figure}[t]
  \centering
  \input{figures/social_network.tex} % generated via DrawFig TikZ export
  \caption{Community structure discovered in the co-authorship graph.}
  \label{fig:community_detection}
\end{figure}

Automation ideas - Python emits JSON for each experimental condition - Optional CLI batch rendering when/if exposed by deployment Reproducibility - Diagram sources archived next to code & datasets - Public replication bundle after acceptance

---

Quantitative snapshot (illustrative)

Task	Before (h)	After (h)	Speed-up
One dense network figure	3.0	0.5	~6×
Full set of 8 figures	24	4	~6×
Post-review polish	2.0	0.3	~6.7×
New-hire ramp (days)	~30	~3	~10×

“Figure prep used to be the part I dreaded. Now it is the fastest step in a revision cycle — especially with TikZ export that is cleaner than my hand-written boilerplate.” — composite quote

---

Feature usage mix (fictional analytics table)

Capability	Share
Visual editing	~40%
Structured import (JSON)	~21%
TikZ export	~16%
Templates	~12%
Batch / scripted renders	~8%
Raster / vector exports	~3%

Layouts used most 1. Force-directed 2. Hierarchical 3. Circular 4. Grid Palettes 1. Academic 2. Colour-blind safe 3. Vibrant 4. Monochrome

---

Tips from the “lab playbook”

Tip 1 — Script the graph, polish in DrawFig

import json
import networkx as nx

G = nx.barabasi_albert_graph(50, 3)
payload = {
    "nodes": [{"id": n, "label": str(n)} for n in G.nodes()],
    "edges": [{"source": u, "target": v} for u, v in G.edges()],
}
with open("network.json", "w", encoding="utf-8") as fh:
    json.dump(payload, fh, indent=2)
# Import JSON → layout → manual nudge labels & colours

Tip 2 — Clone a template per paper section

Start from the “social network analysis” starter, swap labels, retune forces.

Tip 3 — Treat automation as optional glue

configs/network_small.json   # 50 nodes
configs/network_medium.json  # 100 nodes
# Hypothetical batch helper — availability depends on product build
# drawfig batch-render configs/ --output figures/ --format pdf

Tip 4 — Repository layout

drawfig-figures/
  paper1/
    figure1.drawio
    exported/
      figure1.pdf
      figure1.tex
  templates/
    social-network.drawio

---

Road map ideas from the team

1. Custom plugins for domain glyphs (neurons, regulatory motifs)
2. Publish their tuned templates under an open licence
3. Product feedback: temporal networks, inline graph metrics, Jupyter bridges

---

Advice for other labs

Do

1. Write a one-page style guide (fonts, sizes, colours)
2. Version-control diagram sources with Git
3. Start from templates, then fork per project
4. Prefer data import over retyping adjacency lists
5. Archive PDF + TikZ + editable source before camera-ready

Don’t

1. Trust auto-layout as the final word — always eyeball crossings & labels
2. Ignore export DPI / font embedding / crop boxes
3. Mix incompatible editor builds inside one team
4. Keep only bitmaps — you will need the vector/TikZ source again

---

Takeaways

• Throughput: hours → tens of minutes for comparable network figures
• Quality: consistent styling that survived external review
• Training: juniors productive in days instead of months
• Reproducibility: LaTeX-friendly exports + Git-tracked sources

“If we had adopted this earlier, thesis timelines would have shifted by months. Recommend it to anyone publishing graph-heavy CS work.” — composite quote

---

Related reading

• Graph drawing primer
• Network diagram guide
• DrawFig vs FigDraw vs BioRender
• TikZ deep dive (site): /blog/tikz-export-tutorial

Want to share your story? Email contact@drawfig.com with a short outline.

---

Composite narrative with anonymised metrics — authorised for marketing use. Updated 2026-03-10