paper2tex

📄 paper2tex

Convert your Word paper into conference-formatted LaTeX in seconds — with 0 corrupted math, 0 question marks, and perfectly formatted figures.

---

- The Problem Everyone Faces

Every semester, students and researchers go through the same nightmare:

[Write in Word] ──► [Paste into ChatGPT] ──► [Paste in Overleaf] ──► [3 Hours of Errors]
                         (Tokenizer ruins         (Missing packages,
                          math & ° symbols)        broken tables & figures)

1. ChatGPT re-types your document through its tokenizer, corrupting equations, degree symbols (°), Greek letters, and chemical subscripts (CO2).
2. Figures get squished, rotated sideways, or clustered at the very end of the PDF.
3. You spend hours fighting red LaTeX errors right before the submission deadline.

---

✨ The Solution: paper2tex

paper2tex is an AI agent skill that uses a 100% deterministic local toolchain:

• ✅ Verbatim Text — Zero rewording or "hallucinations". Your exact prose is preserved.
• ✅ Perfect Math — Equations are extracted directly from Microsoft Word's XML equation engine (OMML → MathML → LaTeX).
• ✅ Smart Figures — Automatically handles subfigures (a), (b), (c), auto-corrects rotated/sideways images, and scales them cleanly to column width.
• ✅ Auto-Compile & Fix — Compiles locally, reads the error log, automatically repairs missing packages, and produces a publication-ready PDF.

---

🚀 1-Click Install for AI Agents (Copy & Paste)

Copy and paste this single prompt directly into Antigravity, Claude Code, Cursor, OpenCode, Codex, or Windsurf:

Install the paper2tex skill globally. Clone https://github.com/redfries/paper2tex.git into my global agent skills directory (or ~/.gemini/antigravity/skills/paper2tex, ~/.claude/skills/paper2tex, or current workspace skills), ensure prerequisites are installed (pandoc, tectonic, and python packages: lxml, python-docx, Pillow, pymupdf, requests), and run `python scripts/preflight.py` to verify that paper2tex is ready.

---

⚡ How to Ask Your Agent to Convert a Paper

To make sure your AI uses paper2tex's local toolchain (instead of trying to rewrite the document from memory), use an explicit command:

Use paper2tex skill to convert "paper.docx" into IEEE conference format.

Or for other conferences/journals:

Use paper2tex skill to format "thesis.docx" for ACM SIGCONF, my figures are in ./figures.

Run paper2tex on "manuscript.docx" targeting Springer LNCS.

---

🖼️ How to Provide Figures (The Visual Guide)

paper2tex is designed to handle figures cleanly even if you are non-technical.

1. The Recommended Folder Structure

Place your Word file and an optional figures/ folder in the same place (supports .png, .jpg, .pdf, .svg):

my_paper_project/
│
├── my_paper.docx               <-- Your main paper draft
│
└── figures/                    <-- (Optional folder with your images)
    ├── fig1_a.png              <-- Subfigure (a) (Regular PNG, JPG, or PDF)
    ├── fig1_b.png              <-- Subfigure (b)
    ├── fig1_c.png              <-- Subfigure (c)
    ├── fig2.png                <-- Standalone Figure 2
    └── fig3.pdf                <-- (Optional) Vector PDF plot from Python/Matplotlib

💡 Any format works! You can use everyday .png or .jpg images, or export vector .pdf / .svg plots from Python/MATLAB. Or if you don't have a folder, just say extract and paper2tex gets them right out of Word!

---

2. Single Figures vs. Multi-Panel Subfigures

Here is how to set up your images and captions in Word so the output is flawless:

Case A: Single Figure (1 Image)

[ Your Chart or Image ]

Caption in Word:
Fig. 2: System performance comparison under varied load conditions.

• What paper2tex does: Scales the image to fit the exact column width cleanly and anchors the caption directly underneath.

---

Case B: Multi-Panel Subfigures (a, b, c)

When you have multiple related diagrams in one figure:

┌────────────────────────────────────────┐
│  [ Subfigure A: Greenfield Layout ]    │  <-- (a)
├────────────────────────────────────────┤
│  [ Subfigure B: Partial Reuse Layout ] │  <-- (b)
├────────────────────────────────────────┤
│  [ Subfigure C: Maximal Reuse Layout ] │  <-- (c)
└────────────────────────────────────────┘

Caption in Word:
Fig. 1: The three different scenarios. (a) Case #1: greenfield nuclear power plant; (b) Case #2: reuse of retiring coal plant infrastructure; (c) Case #3: maximal reuse of turbine components.

• What paper2tex does:
  1. Detects (a), (b), (c) automatically.
  2. Splits the main title from each sub-description.
  3. Stacks wide landscape schematics vertically (or places tall charts side-by-side) with official (a), (b), (c) subcaptions beneath each panel.
  4. Rotates any sideways images upright automatically.

---

3. The Upfront Clarification Question

When you tell an AI agent to run paper2tex, the agent will ask:

"Do you have a dedicated figures/ folder with high-resolution/vector images (e.g., ./figures), or should I extract the embedded images directly from the Word document?"

• Simply reply extract: The agent will extract, auto-rotate, and format all embedded images directly from your .docx.
• Or reply with your folder path (e.g., ./figures): The agent will use your high-res vector (.pdf, .svg) or 300+ DPI images for crystal-clear print quality.

---

📋 Supported Conferences & Journals

paper2tex includes built-in recipes for all major academic templates:

Venue / Organization	Template Class	Output Engine	Default Columns
IEEE Conferences	IEEEtran	XeLaTeX	2 Columns
IEEE Transactions & Journals	IEEEtran	XeLaTeX	2 Columns
ACM SIGCONF (CCS, CHI, SIGMOD)	acmart	XeLaTeX	2 Columns
Springer LNCS	llncs	XeLaTeX	1 Column
NeurIPS	neurips	XeLaTeX	1 Column
ICML / CVPR	icml / cvpr	pdfLaTeX / XeLaTeX	2 Columns
Elsevier / MDPI / Nature	Generic / Custom	XeLaTeX	Custom

💡 Have a custom conference template? Just provide the template .zip file from the conference website. paper2tex will automatically extract its .cls file and configure the preamble.

---

🛠️ Manual Installation & CLI (For Developers)

If you prefer to run paper2tex manually from your command line:

1. Install System Tools

# Windows (PowerShell)
winget install JohnMacFarlane.Pandoc
winget install tectonic

# Python Dependencies
pip install lxml python-docx Pillow pymupdf requests

2. Clone & Install

git clone https://github.com/redfries/paper2tex.git
cd paper2tex
pip install -e .

3. Verify System Setup

python scripts/preflight.py

4. Run the Pipeline

# 1. Extract content and figures
python scripts/extract.py paper.docx work/

# 2. Select target template
python scripts/template_spec.py "ieee-conference" work/

# 3. Assemble publication-ready LaTeX
python scripts/assemble.py work/

# 4. Compile to PDF with automatic fix loop
python scripts/compile.py work/main.tex

# 5. Verify character & citation fidelity
python scripts/verify.py work/main.tex

---

❓ Frequently Asked Questions (FAQ)

Will this change or reword my writing?

Never. The #1 core rule of paper2tex is 100% verbatim text preservation. No text is summarized, reworded, or dropped.

Why did my figures look tiny or sideways before?

Microsoft Word stores rotated images with internal rotation angle transforms (rot=5400000). If tools extract the raw file without reading Word's XML transform, the image displays sideways. paper2tex reads the OpenXML transformation tags and normalizes every graphic with Pillow so it appears upright and at full readability.

Can I upload the output to Overleaf?

Yes! paper2tex produces an overleaf.zip inside work/submission/. You can upload this zip file directly into Overleaf (Overleaf → New Project → Upload Project) and compile instantly with 0 errors.

Do I need to install a 6 GB TeX Live distribution?

No. paper2tex uses Tectonic by default — a lightweight, modern TeX engine that automatically downloads only the packages your document requires on the fly (~50 MB).

---

📄 License

MIT License — see LICENSE for details. Open-source and free for all students, researchers, and developers.