This tutorial will show you the use of PyMuPDF, MuPDF in Python, step by step.
Because MuPDF supports not only PDF, but also XPS, OpenXPS, CBZ, CBR, FB2 and EPUB formats, so does PyMuPDF [1]. Nevertheless, for the sake of brevity we will only talk about PDF files. At places where indeed only PDF files are supported, this will be mentioned explicitly.
Importing the Bindings#
The Python bindings to MuPDF are made available by this import statement. We also show here how your version can be checked:
>>> import pymupdf>>> print(pymupdf.__doc__)PyMuPDF 1.16.0: Python bindings for the MuPDF 1.16.0 library.Version date: 2019-07-28 07:30:14.Built for Python 3.7 on win32 (64-bit).
Note on the Name fitz#
Old versions of PyMuPDF had their Python import name as fitz
. Newer versions use pymupdf
instead, and offer fitz
as a fallback so that old code will still work.
The reason for the name fitz
is a historical curiosity:
The original rendering library for MuPDF was called Libart.
“After Artifex Software acquired the MuPDF project, the development focus shifted on writing a new modern graphics library called “Fitz”. Fitz was originally intended as an R&D project to replace the aging Ghostscript graphics library, but has instead become the rendering engine powering MuPDF.” (Quoted from Wikipedia).
Note
Use of legacy name fitz
can fail if defunct pypi.org package fitz
is installed; see Problems after installation.
Opening a Document#
To access a supported document, it must be opened with the following statement:
doc = pymupdf.open(filename) # or pymupdf.Document(filename)
This creates the Document object doc. filename must be a Python string (or a pathlib.Path
) specifying the name of an existing file.
It is also possible to open a document from memory data, or to create a new, empty PDF. See Document for details. You can also use Document as a context manager.
A document contains many attributes and functions. Among them are meta information (like “author” or “subject”), number of total pages, outline and encryption information.
Some Document Methods and Attributes#
Method / Attribute | Description |
---|---|
Document.page_count | the number of pages (int) |
Document.metadata | the metadata (dict) |
Document.get_toc() | get the table of contents (list) |
Document.load_page() | read a Page |
Accessing Meta Data#
PyMuPDF fully supports standard metadata. Document.metadata is a Python dictionary with the following keys. It is available for all document types, though not all entries may always contain data. For details of their meanings and formats consult the respective manuals, e.g. Adobe PDF References for PDF. Further information can also be found in chapter Document. The meta data fields are strings or None if not otherwise indicated. Also be aware that not all of them always contain meaningful data – even if they are not None.
Key | Value |
---|---|
producer | producer (producing software) |
format | format: ‘PDF-1.4’, ‘EPUB’, etc. |
encryption | encryption method used if any |
author | author |
modDate | date of last modification |
keywords | keywords |
title | title |
creationDate | date of creation |
creator | creating application |
subject | subject |
Note
Apart from these standard metadata, PDF documents starting from PDF version 1.4 may also contain so-called “metadata streams” (see also stream). Information in such streams is coded in XML. PyMuPDF deliberately contains no XML components for this purpose (the PyMuPDF Xml class is a helper class intended to access the DOM content of a Story object), so we do not directly support access to information contained therein. But you can extract the stream as a whole, inspect or modify it using a package like lxml and then store the result back into the PDF. If you want, you can also delete this data altogether.
Note
There are two utility scripts in the repository that metadata import (PDF only) resp. metadata export metadata from resp. to CSV files.
Working with Outlines#
The easiest way to get all outlines (also called “bookmarks”) of a document, is by loading its table of contents:
toc = doc.get_toc()
This will return a Python list of lists [[lvl, title, page, …], …] which looks much like a conventional table of contents found in books.
lvl is the hierarchy level of the entry (starting from 1), title is the entry’s title, and page the page number (1-based!). Other parameters describe details of the bookmark target.
Note
There are two utility scripts in the repository that toc import (PDF only) resp. toc export table of contents from resp. to CSV files.
Working with Pages#
Page handling is at the core of MuPDF’s functionality.
You can render a page into a raster or vector (SVG) image, optionally zooming, rotating, shifting or shearing it.
You can extract a page’s text and images in many formats and search for text strings.
For PDF documents many more methods are available to add text or images to pages.
First, a Page must be created. This is a method of Document:
page = doc.load_page(pno) # loads page number 'pno' of the document (0-based)page = doc[pno] # the short form
Any integer -∞ < pno < page_count
is possible here. Negative numbers count backwards from the end, so doc[-1] is the last page, like with Python sequences.
Some more advanced way would be using the document as an iterator over its pages:
for page in doc: # do something with 'page'# ... or read backwardsfor page in reversed(doc): # do something with 'page'# ... or even use 'slicing'for page in doc.pages(start, stop, step): # do something with 'page'
Once you have your page, here is what you would typically do with it:
Inspecting the Links, Annotations or Form Fields of a Page#
Links are shown as “hot areas” when a document is displayed with some viewer software. If you click while your cursor shows a hand symbol, you will usually be taken to the target that is encoded in that hot area. Here is how to get all links:
# get all links on a pagelinks = page.get_links()
links is a Python list of dictionaries. For details see Page.get_links().
You can also use an iterator which emits one link at a time:
for link in page.links(): # do something with 'link'
If dealing with a PDF document page, there may also exist annotations (Annot) or form fields (Widget), each of which have their own iterators:
for annot in page.annots(): # do something with 'annot'for field in page.widgets(): # do something with 'field'
Rendering a Page#
This example creates a raster image of a page’s content:
pix = page.get_pixmap()
pix is a Pixmap object which (in this case) contains an RGB image of the page, ready to be used for many purposes. Method Page.get_pixmap() offers lots of variations for controlling the image: resolution / DPI, colorspace (e.g. to produce a grayscale image or an image with a subtractive color scheme), transparency, rotation, mirroring, shifting, shearing, etc. For example: to create an RGBA image (i.e. containing an alpha channel), specify pix = page.get_pixmap(alpha=True).
A Pixmap contains a number of methods and attributes which are referenced below. Among them are the integers width, height (each in pixels) and stride (number of bytes of one horizontal image line). Attribute samples represents a rectangular area of bytes representing the image data (a Python bytes object).
Note
You can also create a vector image of a page by using Page.get_svg_image(). Refer to this Vector Image Support page for details.
Saving the Page Image in a File#
We can simply store the image in a PNG file:
pix.save("page-%i.png" % page.number)
Displaying the Image in GUIs#
We can also use it in GUI dialog managers. Pixmap.samples represents an area of bytes of all the pixels as a Python bytes object. Here are some examples, find more in the examples directory.
wxPython#
Consult their documentation for adjustments to RGB(A) pixmaps and, potentially, specifics for your wxPython release:
if pix.alpha: bitmap = wx.Bitmap.FromBufferRGBA(pix.width, pix.height, pix.samples)else: bitmap = wx.Bitmap.FromBuffer(pix.width, pix.height, pix.samples)
Tkinter#
Please also see section 3.19 of the Pillow documentation:
from PIL import Image, ImageTk# set the mode depending on alphamode = "RGBA" if pix.alpha else "RGB"img = Image.frombytes(mode, [pix.width, pix.height], pix.samples)tkimg = ImageTk.PhotoImage(img)
The following avoids using Pillow:
# remove alpha if presentpix1 = pymupdf.Pixmap(pix, 0) if pix.alpha else pix # PPM does not support transparencyimgdata = pix1.tobytes("ppm") # extremely fast!tkimg = tkinter.PhotoImage(data = imgdata)
If you are looking for a complete Tkinter script paging through any supported document, here it is!. It can also zoom into pages, and it runs under Python 2 or 3. It requires the extremely handy PySimpleGUI pure Python package.
PyQt4, PyQt5, PySide#
Please also see section 3.16 of the Pillow documentation:
from PIL import Image, ImageQt# set the mode depending on alphamode = "RGBA" if pix.alpha else "RGB"img = Image.frombytes(mode, [pix.width, pix.height], pix.samples)qtimg = ImageQt.ImageQt(img)
Again, you also can get along without using Pillow. Qt’s QImage
luckily supports native Python pointers, so the following is the recommended way to create Qt images:
from PyQt5.QtGui import QImage# set the correct QImage format depending on alphafmt = QImage.Format_RGBA8888 if pix.alpha else QImage.Format_RGB888qtimg = QImage(pix.samples_ptr, pix.width, pix.height, fmt)
Extracting Text and Images#
We can also extract all text, images and other information of a page in many different forms, and levels of detail:
text = page.get_text(opt)
Use one of the following strings for opt to obtain different formats [2]:
“text”: (default) plain text with line breaks. No formatting, no text position details, no images.
“blocks”: generate a list of text blocks (= paragraphs).
“words”: generate a list of words (strings not containing spaces).
“html”: creates a full visual version of the page including any images. This can be displayed with your internet browser.
“dict” / “json”: same information level as HTML, but provided as a Python dictionary or resp. JSON string. See TextPage.extractDICT() for details of its structure.
“rawdict” / “rawjson”: a super-set of “dict” / “json”. It additionally provides character detail information like XML. See TextPage.extractRAWDICT() for details of its structure.
“xhtml”: text information level as the TEXT version but includes images. Can also be displayed by internet browsers.
“xml”: contains no images, but full position and font information down to each single text character. Use an XML module to interpret.
To give you an idea about the output of these alternatives, we did text example extracts. See Appendix 2: Considerations on Embedded Files.
Searching for Text#
You can find out, exactly where on a page a certain text string appears:
areas = page.search_for("mupdf")
This delivers a list of rectangles (see Rect), each of which surrounds one occurrence of the string “mupdf” (case insensitive). You could use this information to e.g. highlight those areas (PDF only) or create a cross reference of the document.
Please also do have a look at chapter Working together: DisplayList and TextPage and at demo programs demo.py and demo-lowlevel.py. Among other things they contain details on how the TextPage, Device and DisplayList classes can be used for a more direct control, e.g. when performance considerations suggest it.
Stories: Generating PDF from HTML Source#
The Story class is a new feature of PyMuPDF version 1.21.0. It represents support for MuPDF’s “story” interface.
The following is a quote from the book “MuPDF Explored” by Robin Watts from Artifex:
Stories provide a way to easily layout styled content for use with devices, such as those offered by Document Writers (…). The concept of a story comes from desktop publishing, which in turn (…) gets it from newspapers. If you consider a traditional newspaper layout, it will consist of various news articles (stories) that are laid out into multiple columns, possibly across multiple pages.
Accordingly, MuPDF uses a story to represent a flow of text with styling information. The user of the story can then supply a sequence of rectangles into which the story will be laid out, and the positioned text can then be drawn to an output device. This keeps the concept of the text itself (the story) to be separated from the areas into which the text should be flowed (the layout).
Note
A Story works somewhat similar to an internet browser: It faithfully parses and renders HTML hypertext and also optional stylesheets (CSS). But its output is a PDF – not web pages.
When creating a Story, the input from up to three different information sources is taken into account. All these items are optional.
HTML source code, either a Python string or created by the script using methods of Xml.
CSS (Cascaded Style Sheet) source code, provided as a Python string. CSS can be used to provide styling information (text font size, color, etc.) like it would happen for web pages. Obviously, this string may also be read from a file.
An Archive must be used whenever the DOM references images, or uses text fonts except the standard PDF Base 14 Fonts, CJK fonts and the NOTO fonts generated into the PyMuPDF binary.
The API allows creating DOMs completely from scratch, including desired styling information. It can also be used to modify or extend provided HTML: text can be deleted or replaced, or its styling can be changed. Text – for example extracted from databases – can also be added and fill template-like HTML documents.
It is not required to provide syntactically complete HTML documents: snippets like <b>Hello
are fully accepted, and many / most syntax errors are automatically corrected.
After the HTML is considered complete, it can be used to create a PDF document. This happens via the new DocumentWriter class. The programmer calls its methods to create a new empty page, and passes rectangles to the Story to fill them.
The story in turn will return completion codes indicating whether or not more content is waiting to be written. Which part of the content will land in which rectangle or on which page is automatically determined by the story itself – it cannot be influenced other than by providing the rectangles.
Please see the Stories recipes for a number of typical use cases.
PDF Maintenance#
PDFs are the only document type that can be modified using PyMuPDF. Other file types are read-only.
However, you can convert any document (including images) to a PDF and then apply all PyMuPDF features to the conversion result. Find out more here Document.convert_to_pdf(), and also look at the demo script pdf-converter.py which can convert any supported document to PDF.
Document.save() always stores a PDF in its current (potentially modified) state on disk.
You normally can choose whether to save to a new file, or just append your modifications to the existing one (“incremental save”), which often is very much faster.
The following describes ways how you can manipulate PDF documents. This description is by no means complete: much more can be found in the following chapters.
Modifying, Creating, Re-arranging and Deleting Pages#
There are several ways to manipulate the so-called page tree (a structure describing all the pages) of a PDF:
Document.delete_page() and Document.delete_pages() delete pages.
Document.copy_page(), Document.fullcopy_page() and Document.move_page() copy or move a page to other locations within the same document.
Document.select() shrinks a PDF down to selected pages. Parameter is a sequence [3] of the page numbers that you want to keep. These integers must all be in range 0 <= i < page_count. When executed, all pages missing in this list will be deleted. Remaining pages will occur in the sequence and as many times (!) as you specify them.
So you can easily create new PDFs with
the first or last 10 pages,
only the odd or only the even pages (for doing double-sided printing),
pages that do or don’t contain a given text,
reverse the page sequence, …
… whatever you can think of.
The saved new document will contain links, annotations and bookmarks that are still valid (i.a.w. either pointing to a selected page or to some external resource).
Document.insert_page() and Document.new_page() insert new pages.
Pages themselves can moreover be modified by a range of methods (e.g. page rotation, annotation and link maintenance, text and image insertion).
Joining and Splitting PDF Documents#
Method Document.insert_pdf() copies pages between different PDF documents. Here is a simple joiner example (doc1 and doc2 being opened PDFs):
# append complete doc2 to the end of doc1doc1.insert_pdf(doc2)
Here is a snippet that splits doc1. It creates a new document of its first and its last 10 pages:
doc2 = pymupdf.open() # new empty PDFdoc2.insert_pdf(doc1, to_page = 9) # first 10 pagesdoc2.insert_pdf(doc1, from_page = len(doc1) - 10) # last 10 pagesdoc2.save("first-and-last-10.pdf")
More can be found in the Document chapter. Also have a look at PDFjoiner.py.
Embedding Data#
PDFs can be used as containers for arbitrary data (executables, other PDFs, text or binary files, etc.) much like ZIP archives.
PyMuPDF fully supports this feature via Document embfile_* methods and attributes. For some detail read Appendix 3, consult the Wiki on dealing with embedding files, or the example scripts embedded-copy.py, embedded-export.py, embedded-import.py, and embedded-list.py.
Saving#
As mentioned above, Document.save() will always save the document in its current state.
You can write changes back to the original PDF by specifying option incremental=True. This process is (usually) extremely fast, since changes are appended to the original file without completely rewriting it.
Document.save() options correspond to options of MuPDF’s command line utility mutool clean, see the following table.
Save Option | mutool | Effect |
---|---|---|
garbage=1 | g | garbage collect unused objects |
garbage=2 | gg | in addition to 1, compact xref tables |
garbage=3 | ggg | in addition to 2, merge duplicate objects |
garbage=4 | gggg | in addition to 3, merge duplicate stream content |
clean=True | cs | clean and sanitize content streams |
deflate=True | z | deflate uncompressed streams |
deflate_images=True | i | deflate image streams |
deflate_fonts=True | f | deflate fontfile streams |
ascii=True | a | convert binary data to ASCII format |
linear=True | l | create a linearized version |
expand=True | d | decompress all streams |
Note
For an explanation of terms like object, stream, xref consult the Glossary chapter.
For example, mutool clean -ggggz file.pdf yields excellent compression results. It corresponds to doc.save(filename, garbage=4, deflate=True).
Closing#
It is often desirable to “close” a document to relinquish control of the underlying file to the OS, while your program continues.
This can be achieved by the Document.close() method. Apart from closing the underlying file, buffer areas associated with the document will be freed.
Further Reading#
Also have a look at PyMuPDF’s Wiki pages. Especially those named in the sidebar under title “Recipes” cover over 15 topics written in “How-To” style.
This document also contains a FAQ. This chapter has close connection to the aforementioned recipes, and it will be extended with more content over time.
Footnotes
This software is provided AS-IS with no warranty, either express or implied. This software is distributed under license and may not be copied, modified or distributed except as expressly authorized under the terms of that license. Refer to licensing information at artifex.com or contact Artifex Software Inc., 39 Mesa Street, Suite 108A, San Francisco CA 94129, United States for further information.
This documentation covers all versions up to 1.24.10.