Convert HTML to PDF in Python

What you need

• An API key. Create one in the app under API keys and export it as FORMFEED_API_KEY. A test key (ff_test_…) is right for trying things out. See the quick start.
• Python. You need Python 3.10 or newer and the formfeed package (it brings httpx and pydantic).
• An HTML file. invoice.html next to the program: a complete document with its CSS inline or linked. A sample is further down.

pip install formfeed

The program

The program reads invoice.html, posts it with the paper size and the margins, waits for the result when the API answers 202, stops with the API’s message when something is wrong, and saves the PDF.

import os
from pathlib import Path

from formfeed import Formfeed

formfeed = Formfeed(api_key=os.environ["FORMFEED_API_KEY"])
html = Path("invoice.html").read_text(encoding="utf-8")

# a problem answer (4xx, 5xx) raises a FormfeedError with the API's code, title and detail
render = formfeed.renders.create(
    html=html,
    settings={
        "paper": {"format": "A4"},
        "margin": {"top": "20mm", "right": "18mm", "bottom": "20mm", "left": "18mm"},
    },
)
# not done within 60 seconds: the API answered 202 and the render goes on in the background
if render.status != "succeeded":
    render = formfeed.renders.wait_for(render.id)
if render.status == "failed":
    raise SystemExit(f"render failed: {render.error}")

Path("invoice.pdf").write_bytes(formfeed.renders.download(render))
print(render.id, render.page_count, "page(s),", render.units, "unit(s)")

Run it

It prints the render’s id, the page count and the units it cost, and invoice.pdf lies next to the program.

export FORMFEED_API_KEY=ff_test_...
python render.py

What is particular about Python

• Keyword arguments. Every field of the request is a keyword argument (renders.create, pdf.protect), after the items of renders.batch and the sources of pdf.merge, which come first; nested options such as settings or post are plain dictionaries.
• Retries are built in. The client sends an Idempotency-Key with every render and retries 429, 503 and network errors up to three times with backoff, with the same key: a retry of a request that was answered already gets that answer back instead of a second render.
• Typed results. The answer is a pydantic model: render.download_url, render.page_count and render.units are attributes, not dictionary keys.
• Async too. AsyncFormfeed has the same methods as coroutines, for FastAPI or any asyncio service.

An HTML file that prints well

The HTML is printed as it is, with the print rules Chromium supports. Web fonts load from a <link>, and two CSS rules keep a long table readable across pages:

<!doctype html>
<html lang="en">
  <head>
    <meta charset="utf-8">
    <link rel="stylesheet" href="https://fonts.googleapis.com/css2?family=Inter:wght@400;700&display=swap">
    <style>
      body { font-family: 'Inter', sans-serif; font-size: 10pt; color: #1e293b; }
      h1 { font-size: 20pt; color: #312e81; }
      table { width: 100%; border-collapse: collapse; }
      td, th { padding: 2mm; border-bottom: 1px solid #e2e8f0; text-align: left; }
      thead { display: table-header-group; } /* repeats the head row on every page */
      tr { break-inside: avoid; }
    </style>
  </head>
  <body>
    <h1>Invoice 2026-0042</h1>
    <table>
      <thead><tr><th>Item</th><th>Amount</th></tr></thead>
      <tbody><tr><td>Consulting</td><td>960.00 €</td></tr></tbody>
    </table>
  </body>
</html>

What the settings control

• Paper. paper.format takes A4, Letter and the other common names, landscape: true turns it, and width with height sets a custom size such as a 100 × 150 mm label.
• Margins. Per side, as CSS lengths such as 20mm. A background that should reach the edge needs margins of 0.
• Header and footer. header.html and footer.html repeat on every page inside the top and bottom margin; elements with the classes pageNumber and totalPages show the page numbers.
• Placeholders. Add "engine": "jinja2" (or liquid, handlebars) and a data object, and {{ … }} in the HTML is filled before printing, with all helpers for money, dates, barcodes and charts.

When a template is the better tool

Sending HTML is right for pages your application already renders. When the document is yours to design (invoices, certificates, reports), a stored template gives you an editor with a page-accurate preview, versions, and a request that carries only data. The PDF invoice guide of this language shows that variant.

When the API says no

• One shape for every error. Errors are problem documents (RFC 9457) with type, title, status, detail, a stable code and the request_id to quote to support; the type URL leads to a page about that error. The list is in the error reference.
• The usual ones. 401 unauthorized for a missing or revoked key, 404 template_not_found for a wrong template name, 422 template_runtime_error when the template fails on the data, 429 rate_limited above the plan’s requests per second, 402 quota_exceeded when the Free plan’s units are used up (a paid plan bills extra units up to its hard cap).
• Retry safely. Send an Idempotency-Key header with the POST, and repeat the request with the same key when you do not know whether the first attempt arrived: for 24 hours the key returns the first answer instead of rendering and billing again. The answer is stored once the first attempt has been answered, so do not retry while it may still run; and problems below 500 are stored too, so a corrected request needs a new key.

The job around this program

This guide is one program. What the job needs around it, from the template to the limits and the price, is on the page HTML to PDF API.

More Python guides

• Generate a PDF invoice from JSON in Python
• Create a ZUGFeRD / Factur-X e-invoice in Python
• Generate Open Graph images in Python
• Generate PDFs from a CSV in Python
• Merge PDFs and add a password in Python
• Async PDF rendering with a verified webhook in Python

FAQ

Can it render a URL instead of an HTML string?

Yes: send url instead of html. It must be a public https page, which is loaded without your cookies or headers; addresses in private networks are refused, and a page that does not load within 30 seconds fails with navigation_failed.

How large may the HTML be?

Up to 5 MB per request. Images and fonts are fetched by URL and do not count towards that. Limits per plan are in the docs.

Does JavaScript in the page run?

Yes, Chromium runs it before printing, which is how charts render. It waits until the network has been idle, for at most 30 seconds, so a page that keeps loading fails with render_timeout rather than hanging.

HTML to PDF in other languages

• Node.js (TypeScript)
• PHP
• Go
• C# (.NET)
• Java
• Ruby
• curl