LaTeXtify

Convert scientific manuscripts from Word (.docx) into journal-ready LaTeX projects and PDFs. Pick a journal, point at a manuscript, get a compilable LaTeX project with extracted citations, journal-appropriate formatting, and a compiled PDF — no TeX installation required (the Tectonic engine is downloaded and managed automatically).

latextify convert paper.docx --journal revtex4-2 --pdf

Try it online: https://latextify-demo.onrender.com — a free-hosted demo of the GUI (shared low-power hardware: conversions are slow, uploads capped at 25 MB, rate-limited, and the instance takes ~1 min to wake after idle). Do not upload confidential or unpublished manuscripts to the demo; install locally for private, full-speed conversions.

What it does

• Journal formatting — emits a project using the journal's real document class with correct author/affiliation macros, compiled with Tectonic. Class files missing from Tectonic's bundle are vendored (where licenses permit) and staged automatically.
• Citations with links — extracts references from Zotero/Mendeley field codes, EndNote, Word's built-in citation manager, or hand-typed text (reconstructed via Crossref with confidence scoring), emits references.bib with DOIs, and links in-text citations to real \cite{} commands. Citation style (numeric ↔ author-year) switches per journal support.
• Reference checking + review (opt-in, needs internet) — --check-references validates every reference against Crossref: it resolves each DOI and compares the stored title, authors, year, journal, volume, and pages against the canonical record, flags dead DOIs, and suggests DOIs for references that lack one. Findings land in report.md; a dropped connection marks references unchecked rather than failing the run. Add --review (or the GUI's review panel) to step through each flagged reference one by one — approve Crossref's fix, keep yours, or edit the whole entry — then the accepted corrections are written into references.bib and the PDF is recompiled. On by default in the GUI.
• Figures — extracts embedded images, or swap in your own vector/high-res files by dropping figures/fig1.pdf beside the docx (or an explicit figures.yaml). SVG converts to PDF automatically. latextify figures paper.docx lists what you have and what it will print at — see Replacing pasted screenshots with vector art. Embedded camera metadata — GPS, body/lens serial numbers, the photographer's name, the capture thumbnail — is stripped on the way into figures/, losslessly: pixels and colour profile are untouched. --keep-figure-metadata opts out.
• Equations — Word equation editor (OMML) math converts to LaTeX; latextify equations produces a side-by-side audit for equation-heavy papers.
• Tables — clean Word tables become booktabs; pathological merged-cell tables degrade to a compilable, clearly-marked simplification instead of silent corruption.
• Honest by design — every run writes report.md: preflight findings, per-reference citation confidence with "verify me" flags, figure provenance, compile diagnostics. The quality bar is compiles cleanly + punch list, not silent camera-ready claims.
• Re-run safe — main.tex is written once and never overwritten; your manual LaTeX polish survives re-conversion (regenerated content lives in generated/).

Replacing pasted screenshots with vector art

A figure pasted into Word as a screenshot is pixels forever. Printed at column width it is usually well under the 300 DPI journals ask for, and no conversion step can recover detail that was never captured. The fix is to supply the original vector file — LaTeXtify will use it instead of what is in the document.

1. See what you actually have

latextify figures paper.docx

3 figure(s) in paper.docx

FIG  SOURCE    KIND        SIZE           DPI       CAPTION
  1  embedded  raster      1100x900       324       Device schematic and measurement setup.
  2  embedded  raster      640x480        91 LOW    Raman spectra of the three samples.
  3  override  raster-pdf  468x351pt      109 LOW   Temperature dependence of the resistance.

0 of 3 already vector.
Figure(s) 2, 3 would print below 300 DPI. Supply a vector version of each as figures/fig<N>.pdf (or .eps/.svg) beside the manuscript.

Three things this tells you that guessing cannot:

• Which figure is number 3. Figures are numbered by position in the document, which is not always what a caption says — a caption with no image behind it shifts every later number, and the listing warns when it finds one. The number is what names the replacement file, so getting it wrong silently swaps two figures in your submission.
• What each file really is. raster-pdf above means a PDF whose content is a single image and no text: a screenshot someone printed to PDF. It has a vector file's extension and a raster's limits — note that figure 3 is an override, i.e. a file someone supplied believing it was vector.
• How it will print. DPI is the effective resolution at the width it will be printed at, flagged LOW below 300. A landscape figure is assumed to span both columns, so it needs roughly twice the pixels of a single-column one — which is why figure 2 above fails at 640x480 while figure 1 passes at 1100x900. Any crop you applied in Word is accounted for, since those pixels never reach the output.

The same table is in the web GUI under What are my figures?, and --json makes it scriptable.

2. Drop the vector files in

Put them beside the manuscript, named by figure number:

paper.docx
figures/
  fig2.pdf
  fig3.svg

PDF, EPS and SVG are all accepted; SVG and EPS are converted to PDF for you. A vector file always beats the embedded image, and beats a raster of the same number. For names that don't follow the convention, map them explicitly in a figures.yaml beside the docx:

2: plots/raman-spectra.pdf
3: plots/resistance-vs-temperature.pdf

Or keep them wherever they already live:

# a folder anywhere, same fig<N> naming
latextify convert paper.docx -j revtex4-2 --figures-dir ~/plots

# one file, named outright (beats every other source)
latextify convert paper.docx -j revtex4-2 --figure 3=~/plots/spectra.pdf

# everything in one exported PDF: page N replaces figure N
latextify convert paper.docx -j revtex4-2 --figures-pdf ~/plots/all-figures.pdf

latextify figures takes the same three options, so you can preview exactly what a conversion would use before running it.

In the web GUI, drop the bundle in as a figure bundle (PDF) rather than as an ordinary figure — as a plain figure only its first page would be used. An individually uploaded figure still wins over that number's page in the bundle.

A figure you never pasted into Word

If the manuscript captions Figure 3 but no image was ever pasted for it, supplying fig3.pdf inserts it at that caption. This is worth doing even if you don't care about vector art, because of what the gap does otherwise:

Word binds a caption to the image next to it, so a missing figure 3 means the image belonging to caption 4 gets emitted carrying caption 3's text, and every later figure is shifted the same way. The orphaned "Figure 3" caption is left behind in the body as ordinary prose. Nothing about the output looks broken; the figures are simply mislabelled.

Supplying the missing file repairs all of it — the figures are renumbered to the numbers your captions state, so each image is re-paired with its own caption, and the new figure is planted where the orphan paragraph sat. Without a replacement file nothing is renumbered; you get the warning naming the fix, exactly as before.

3. Convert, and check it took

latextify convert paper.docx -j revtex4-2 --pdf --vector-figures

--vector-figures reports every figure that is still raster, with its print DPI and the exact filename to supply:

warning: figure 3 appears to be a raster image wrapped in a PDF -- about 109 DPI
at 7 in wide, below the 300 DPI most journals require. Supply a vector version
as figures/fig3.pdf (PDF, EPS or SVG) to replace it.

report.md records the same thing durably, per figure:

**Fig 1** (EMBEDDED, raster 324 DPI at 3.4 in)
**Fig 2** (EMBEDDED, raster 91 DPI at 7 in — below 300 DPI)
**Fig 3** (OVERRIDE, raster in a PDF 109 DPI at 7 in — below 300 DPI)

Getting vector out of Word in the first place

Word is not the problem as often as it seems. It cannot hold vector for a screen capture — that is raster the moment you press Print Screen — but it holds real vector for anything drawn:

• Paste Special, not Ctrl+V. Copy a chart from Excel, Origin, MATLAB or Illustrator, then Home ▸ Paste ▸ Paste Special ▸ Picture (Enhanced Metafile). That lands as EMF, which is vector. A plain Ctrl+V usually gives you a bitmap.
• LaTeXtify converts EMF and WMF to PDF when LibreOffice or Inkscape is on your PATH or in its usual Windows install directory. If neither is available, it tries a 600-DPI PNG fallback and warns that the result is no longer vector; a metafile it cannot safely decode is skipped with an actionable warning.
• Export from the plotting tool. Saving directly to PDF or SVG and dropping it in as figures/fig<N>.pdf skips Word entirely and is the most reliable route.
• A micrograph or photograph has no vector form. Supply it at a size that reaches 300 DPI at print width instead — roughly 1000 px across for a single-column figure, 2100 px for one spanning both columns. latextify figures reports exactly this, so you can check before submitting rather than after a desk rejection.

Supported journals

Fourteen journal templates covering the major physics/chemistry publishers. The -j/--journal value is the short id in the first column; the GUI shows the full publisher name.

The APS and AIP entries are REVTeX variants — the same revtex4-2 class with the publisher's society options and bibliography style, so they compile with no extra class files.

¹ Wiley's class file is proprietary and cannot be redistributed; the template works once you place WileyNJD-v2.cls in the output directory (the error message tells you exactly what is missing).

Adding a journal is data, not code: a folder with a manifest and two Jinja templates. See CONTRIBUTING.md.

Install

Install from PyPI with pip (Python 3.10+):

pip install latextify

That is the whole install: no TeX setup (the Tectonic engine is fetched and managed automatically the first time you use --pdf), and pandoc ships bundled in the wheel. The latextify command is then on your PATH. Also installable with uv pip install latextify or pipx install latextify.

For the drag-and-drop web GUI, add the gui extra:

pip install "latextify[gui]"

From source (development)

git clone https://github.com/pquarterman17/LaTeXtify
cd LaTeXtify
uv sync

One-click launcher

To start the web GUI without touching a command line, double-click startup.bat (Windows) or run ./startup.sh (macOS/Linux). It installs dependencies on first run, opens the GUI in your browser, and — if anything goes wrong — writes an easy-to-find latextify-startup.log next to the script and prints it for you to copy. This source-tree launcher needs internet access on its first run.

Offline / firewalled Windows computer

Do not transfer the repository's source-code ZIP: its startup.bat must fetch uv and Python packages. On a connected computer, download the release asset named latextify-offline-windows-x64.zip, transfer it using an approved method, extract it to a short writable path such as C:\LaTeXtify, and run:

py install.py

Then double-click Start-LaTeXtify-GUI.bat. The offline package contains the GUI, Pandoc, Tectonic, dependencies, and a warmed TeX cache; it does not contact GitHub, Astral, PyPI, or Crossref. It requires a supported 64-bit Python already present on the work computer. Embedded EMF/WMF figures work, but preserving them as vectors requires an offline installation of LibreOffice or Inkscape; otherwise Windows attempts a 600-DPI raster fallback.

Usage

Installed from source with uv sync? Prefix these with uv run (e.g. uv run latextify convert ...), or activate the project venv first.

# Convert + compile to PDF (report.md written alongside)
latextify convert paper.docx --journal revtex4-2 --pdf

# Choose citation style where the journal supports both
latextify convert paper.docx --journal elsarticle --citation-style authoryear

# Supplementary material: S-numbered second document sharing the bibliography
latextify convert paper.docx --journal revtex4-2 --supplement si.docx --pdf

# Main text + supplement in ONE Word file and ONE main.pdf. Put a heading named
# "Supplementary Material", "Supplementary Information", "Supplemental
# Material", or "Supporting Information" at the boundary.
latextify convert merged.docx --journal revtex4-2 --inline-supplement --pdf

# Inline supplements cannot be combined with --figures-at-end: delayed floats
# would cross the numbering boundary. LaTeXtify rejects that combination.
# If the merged file contains separate main and supplementary reference lists,
# LaTeXtify combines them into one bibliography at the end of the PDF. Numeric
# citation numbering is continuous across the whole document and a warning is
# written to the conversion report so this policy is never implicit.

# ...and staple the main text + supplement into one combined.pdf
latextify convert paper.docx --journal revtex4-2 --supplement si.docx --pdf --combine-supplement

# ...or render the supplement as a simplified one-column article (common for SI,
# where formatting rules are looser), keeping the shared bibliography + S-numbers
latextify convert paper.docx --journal revtex4-2 --supplement si.docx --pdf --supplement-onecolumn

# Validate references online against Crossref (DOIs, titles, authors, years...)
latextify convert paper.docx --journal revtex4-2 --check-references

# ...and interactively approve/deny/edit each correction, then rewrite + recompile
latextify convert paper.docx --journal revtex4-2 --pdf --review

# List a manuscript's figures: number, caption, format, effective print DPI
latextify figures paper.docx

# Convert, reporting every figure that is still a screenshot rather than vector
latextify convert paper.docx --journal revtex4-2 --pdf --vector-figures

# Force selected embedded or external figures to one/two columns. Numbers are
# document order as shown by `latextify figures`; all others stay automatic.
latextify convert paper.docx --journal revtex4-2 --pdf \
  --figure-column 1=one --figure-column 2=two

# A folder of manuscripts at once (continue-on-error + summary)
latextify batch drafts/ --journal revtex4-2 --pdf

# Equation conversion audit for equation-heavy papers
latextify equations paper.docx --pdf

# Local web GUI (drag-and-drop; requires the gui extra: pip install "latextify[gui]")
latextify gui

# List registered journals and their citation modes
latextify journals

Web GUI

latextify gui starts a local, browser-based front end (bound to 127.0.0.1 only — your uploads never leave your machine) and opens a tab. Drop your whole submission in at once — main .docx, supplement .docx, figure files, and a .bib reference library together — then set each file's role, pick a journal from the full publisher list, choose options, and click Preview. The GUI reads the selected manuscript's figures and gives every figure its own Automatic / 1 column / 2 columns selector; separately uploaded replacements appear under the same document-order number. The compiled PDFs render inline so you can confirm the conversion worked. A preview is held only in temporary local storage (a private working directory the app owns) — it is pruned automatically about an hour after its last use and deleted when the app shuts down, so nothing lingers unless you keep it. Once it looks good, the Export panel lets you pick a destination folder (a native "Browse…" dialog) and copy any subset of the outputs — the LaTeX project, individual PDFs, or the .zip — to the folder you choose to keep.

For a merged main+supplement Word file, upload it only as Main text, do not add a separate Supplement file, and choose one Combined main + supplement Word file layout:

• Single-column throughout (closest to Word)
• Main two columns; supplement two columns
• Main two columns; supplement one column

Every combined mode produces one main.pdf, starts the recognized Supplementary Material/Information heading on a new page, and restarts figures, tables, equations, and sections as S1. Check references online, Strip figure metadata, and Optimize figure placement are independent toggles. The offline launcher disables the online reference check; the other two remain available. Manual per-figure column choices override automatic placement.

Output layout per conversion:

output/<journal>/
├── main.tex          # yours — written once, never overwritten on re-runs
├── generated/        # regenerated every run: preamble, metadata, body, bibliography
├── figures/          # resolved figure files (embedded or your overrides)
├── references.bib    # extracted bibliography with DOIs
└── report.md         # what happened, what to verify

On first conversion a paper.yaml sidecar is written beside your docx with the guessed title/authors/affiliations — correct it once; it is the source of truth afterwards.

Examples

Three runnable, self-contained examples live in examples/ — each generates its own Word document(s) (no committed binaries) and converts them to a PDF with a single python run.py:

• 01 – all-embedded: one .docx with embedded figures and a typed reference list (Crossref reconstruction).
• 02 – word-plus-figures: external figure files via the figures/ folder convention and a figures.yaml manifest.
• 03 – multipart-refmanager: main + --supplement documents with Zotero/Mendeley field-code citations and a shared, de-duplicated bibliography.

Input expectations

LaTeXtify targets manuscripts that use Word styles: styled headings, equation-editor math, inline figures with captions. Unsupported constructs (text boxes, SmartArt, tracked changes) are reported by preflight rather than silently mangled.

Development

uv run pytest                                     # full suite (~875 tests, real docx→PDF compiles)
uv run pytest -m "not tectonic and not network"   # fast subset
uv run ruff check .

See CONTRIBUTING.md and the archived build plan under plans/archive/ for architecture documentation.

License

Apache-2.0 (see LICENSE). Vendored LaTeX class/style files under latextify/templates/journals/*/vendor/ remain under their own licenses (LPPL) — see NOTICE and the per-journal VENDOR_LICENSE.txt files.