weasyprint-pdf-creation

WeasyPrint PDF Creation

Overview

Use this skill to create polished PDFs from source-controlled HTML/CSS templates. The core workflow is:

HTML template + print CSS + data → WeasyPrint → PDF → QA → publish/share

This is for new PDFs you own: invoices, reports, proposals, briefs, statements, one-page docs, printable guides, and template-driven business documents.

Do not use this as the default for editing arbitrary existing PDFs or capturing JavaScript-heavy webpages. See the decision table below.

When to Use

Use this skill when the user asks to:

• create an invoice, report, proposal, brief, contract-style doc, handout, or printable PDF
• make a reusable PDF template from HTML/CSS
• generate a better-looking PDF than a markdown/export default
• publish a PDF preview for review on an artifact shelf/static host
• test whether a designed document survives print/PDF rendering

Do not use this skill for:

• editing text in an existing PDF → use nano-pdf/PyMuPDF
• OCR/extracting text from scans → use OCR/document extraction tooling
• screenshots or JS-rendered dashboards → use Playwright/Chromium
• raw programmatic vector drawing → consider ReportLab/SVG/PyMuPDF

Decision Table

Need	Tool
New designed invoice/report/doc from template	WeasyPrint
Existing PDF text edit	nano-pdf / PyMuPDF
JS-heavy webpage or dashboard to PDF	Playwright/Chromium
Extract/OCR a PDF	OCR/document tools
Low-level PDF surgery/merge/split	pikepdf / PyMuPDF

Recommended Repo Shape

pdf-project/
  templates/
    invoice.html
    report.html
    proposal.html
    one-page-brief.html
    receipt.html
    statement.html
  styles/
    print.css
  data/
    sample-invoice.json
    sample-report.json
    sample-proposal.json
  scripts/
    render_pdf.py
    qa_pdf.py
  outputs/
    .gitkeep
  examples/
    index.html
    invoice.pdf
    invoice.png
  docs/
    pdf-qa.md
    template-rules.md
  fonts/
    README.md
  requirements.txt
  README.md

Keep templates boring to run and tasteful to look at:

• semantic HTML
• print-first CSS
• no required JavaScript
• local assets/fonts where possible
• data injection separated from layout
• source-controlled sample JSON
• deterministic generated output

Install

Python venv

python3 -m venv .venv
. .venv/bin/activate
pip install -r requirements.txt
python -m weasyprint --info

requirements.txt minimum:

weasyprint==68.1
Jinja2>=3.1
pypdf>=5.0

Debian/Ubuntu native dependencies

For most modern Debian/Ubuntu hosts:

sudo apt-get update
sudo apt-get install -y \
  python3-pip \
  libpango-1.0-0 \
  libpangoft2-1.0-0 \
  libharfbuzz-subset0 \
  fontconfig \
  fonts-dejavu-core

Template Rules

CSS

Use @page and print media deliberately:

@page { size: A4; margin: 0; }
@media print {
  body { background: white; }
  .page { width: 210mm; height: 297mm; overflow: hidden; }
  .no-print { display: none; }
}

Prefer:

• mm for page canvas and key print spacing
• CSS grid/table for stable document layout
• font-variant-numeric: tabular-nums for money and IDs
• explicit column widths for invoice tables
• reserved space for stamps/badges/decorative elements
• wrapping rules for long strings like IBANs, emails, IDs

Avoid or verify carefully:

• absolute positioning near content
• unsupported browser-only CSS
• fancy effects like box-shadow, filter, mix-blend-mode
• JS-only chart/render flows
• content that fits only because sample text is short

HTML

Use semantic blocks and stable classes:

<section class="page invoice">
  <header class="invoice-header">...</header>
  <section class="invoice-meta">...</section>
  <table class="line-items">...</table>
  <section class="payment-box">...</section>
</section>

For long values:

.long-value { overflow-wrap: anywhere; }
.amount { white-space: nowrap; text-align: right; }

Render Script Pattern

Use a small Python script so rendering is repeatable and testable.

from pathlib import Path
import argparse, json
from jinja2 import Environment, FileSystemLoader, select_autoescape
from weasyprint import HTML

parser = argparse.ArgumentParser()
parser.add_argument('--template', required=True)
parser.add_argument('--data', required=True)
parser.add_argument('--out', required=True)
args = parser.parse_args()

template_path = Path(args.template)
data = json.loads(Path(args.data).read_text())

env = Environment(
    loader=FileSystemLoader(str(template_path.parent)),
    autoescape=select_autoescape(['html', 'xml'])
)
html = env.get_template(template_path.name).render(**data)
Path(args.out).parent.mkdir(parents=True, exist_ok=True)
HTML(string=html, base_url=str(template_path.parent.resolve())).write_pdf(args.out)

QA Checklist

Run all of these before calling a PDF done.

1. WeasyPrint warning check

python scripts/render_pdf.py --template templates/invoice.html --data data/sample-invoice.json --out outputs/invoice.pdf 2> outputs/weasyprint.log
test ! -s outputs/weasyprint.log

Warnings often reveal unsupported CSS or broken assets. Fix them unless they are intentionally harmless and documented.

2. PDF metadata/page check

pdfinfo outputs/invoice.pdf | grep -E 'Pages|Page size|PDF version'

Expected A4:

Page size: 595.276 x 841.89 pts (A4)

3. Text extraction smoke test

pdftotext outputs/invoice.pdf - | grep -E 'Invoice|Total|Payment'

This catches invisible text, missing content, and many accidental blank-page failures.

4. Browser preview

Open the HTML preview and check console errors. For artifact shelf/static hosting:

artifact add-dir my-pdf-lab ./public --title "PDF Lab" --type site --tag artifact
curl -fsSI https://artifacts.example.com/artifacts/my-pdf-lab/

5. Visual/overlap check

Inspect the rendered HTML/PDF visually. Focus on:

• totals and money columns
• payment boxes
• long IDs/IBAN/email fields
• stamps/badges
• footer spacing
• page breaks
• table rows at page boundaries

If the user reports overlap, do not hand-wave it. Re-check with browser/vision and redesign the cramped area.

Publishing Pattern

For review, publish a static folder containing previews and PDFs:

public/
  index.html
  invoice.html
  invoice.pdf
  report.html
  report.pdf

Then:

artifact add-dir pdf-lab public --title "PDF Lab" --type site --tag artifact
curl -fsSI https://artifacts.omarbenaidy.com/artifacts/pdf-lab/

If using a different environment, any static host works: GitHub Pages, Cloudflare Pages, Netlify, nginx, S3, etc.

Font Guidance

Fonts affect pagination and overflow. For reproducible PDFs:

• define an explicit font stack
• bundle or install production fonts
• avoid relying on macOS-only fonts unless the renderer runs on macOS
• re-run QA after font changes
• keep fonts/README.md updated with chosen fonts and licenses

Common Pitfalls

1. Treating browser CSS support as WeasyPrint CSS support. WeasyPrint is excellent for print CSS, but not a full Chrome clone. Generate once early and read warnings.

2. Decorative absolute positioning over real content. Stamps, blobs, and badges need reserved layout space. If they float over payment/totals/footer, they will overlap with real data.

3. Long values in single-line fields. IBANs, addresses, emails, and invoice references need overflow-wrap:anywhere or stacked layout.

4. Accidental extra blank pages. Fixed min-height:297mm plus padding/content can overflow. In print mode, use height:297mm, controlled padding, and inspect page count with pdfinfo.

5. Unpinned renderer versions. Rendering can change across WeasyPrint versions. Pin the version and re-check golden PDFs after upgrades.

6. Unpinned fonts. Missing or different fonts change layout. Install known fonts and avoid relying on system-specific typography.

7. Screenshot-only QA. Screenshots are secondary. Check logs, page count, extracted text, browser console, and actual PDF availability.

Verification Checklist

• WeasyPrint version pinned
• Native dependencies/fonts installed and documented
• Render command is repeatable
• Render log has zero unexpected warnings
• pdfinfo page count/size is expected
• pdftotext finds key content
• Browser preview has no console errors
• Visual QA checks totals/payment/footer/page breaks
• Public/review links return 200 when publishing is requested