This class represents a document. It can be constructed from a file or from memory.

There exists the alias open for this class, i.e. fitz.Document(...) and do exactly the same thing.

For details on embedded files refer to Appendix 3.


Starting with v1.17.0, a new page addressing mechanism for EPUB files only is supported. This document type is internally organized in chapters such that pages can most efficiently be found by their so-called “location”. The location is a tuple (chapter, pno) consisting of the chapter number and the page number in that chapter. Both numbers are zero-based.

While it is still possible to locate a page via its (absoute) number, doing so may mean that the complete EPUB document must be layouted before the page can be addressed. This may have a significant performance impact if the document is very large. Using the page’s (chapter, pno) prevents this from happening.

To maintain a consistent API, PyMuPDF supports the page location syntax for all file types – documents without this feature simply have just one chapter. Document.load_page() and the equivalent index access now also support a location argument.

There are a number of methods for converting between page numbers and locations, for determining the chapter count, the page count per chapter, for computing the next and the previous locations, and the last page location of a document.

Method / Attribute Short Description
Document.add_layer() PDF only: make new optional content configuration
Document.add_ocg() PDF only: add new optional content group
Document.authenticate() gain access to an encrypted document
Document.can_save_incrementally() check if incremental save is possible
Document.chapter_page_count() number of pages in chapter
Document.close() close the document
Document.convert_to_pdf() write a PDF version to memory
Document.copy_page() PDF only: copy a page reference
Document.del_toc_item() PDF only: remove a single TOC item
Document.delete_page() PDF only: delete a page
Document.delete_pages() PDF only: delete multiple pages
Document.embfile_add() PDF only: add a new embedded file from buffer
Document.embfile_count() PDF only: number of embedded files
Document.embfile_del() PDF only: delete an embedded file entry
Document.embfile_get() PDF only: extract an embedded file buffer
Document.embfile_info() PDF only: metadata of an embedded file
Document.embfile_names() PDF only: list of embedded files
Document.embfile_upd() PDF only: change an embedded file
Document.extract_font() PDF only: extract a font by xref
Document.extract_image() PDF only: extract an embedded image by xref
Document.ez_save() PDF only: with different defaults
Document.find_bookmark() retrieve page location after layouting document
Document.fullcopy_page() PDF only: duplicate a page
Document.get_layer() PDF only: lists of OCGs in ON, OFF, RBGroups
Document.get_layers() PDF only: list of optional content configurations
Document.get_oc() PDF only: get OCG /OCMD xref of image / form xobject
Document.get_ocgs() PDF only: info on all optional content groups
Document.get_ocmd() PDF only: retrieve definition of an OCMD
Document.get_page_fonts() PDF only: list of fonts referenced by a page
Document.get_page_images() PDF only: list of images referenced by a page
Document.get_page_labels() PDF only: list of page label definitions
Document.get_page_numbers() PDF only: get page numbers having a given label
Document.get_page_xobjects() PDF only: list of XObjects referenced by a page
Document.get_toc() extract the table of contents
Document.get_page_pixmap() create a pixmap of a page by page number
Document.get_page_text() extract the text of a page by page number
Document.get_sigflags() PDF only: determine signature state
Document.get_xml_metadata() PDF only: read the XML metadata
Document.has_annots() PDF only: check if PDF contains any annots
Document.has_links() PDF only: check if PDF contains any links
Document.insert_page() PDF only: insert a new page
Document.insert_pdf() PDF only: insert pages from another PDF
Document.journal_enable() PDF only: enables journalling for the document
Document.journal_start_op() PDF only: start an “operation” giving it a name
Document.journal_stop_op() PDF only: end current operation
Document.journal_position() PDF only: return journalling status
Document.journal_op_name() PDF only: return name of a journalling step
Document.journal_can_do() PDF only: which journal actions are possible
Document.journal_undo() PDF only: undo current operation
Document.journal_redo() PDF only: redo current operation
Document.journal_save() PDF only: save joural to a file
Document.journal_load() PDF only: load joural from a file
Document.layer_ui_configs() PDF only: list of optional content intents
Document.layout() re-paginate the document (if supported)
Document.load_page() read a page
Document.make_bookmark() create a page pointer in reflowable documents
Document.xref_xml_metadata() PDF only: xref of XML metadata
Document.move_page() PDF only: move a page to different location in doc
Document.need_appearances() PDF only: get/set /NeedAppearances property
Document.new_page() PDF only: insert a new empty page
Document.next_location() return (chapter, pno) of following page
Document.outline_xref() PDF only: xref a TOC item
Document.page_cropbox() PDF only: the unrotated page rectangle
Document.pages() iterator over a page range
Document.page_xref() PDF only: xref of a page number
Document.pdf_catalog() PDF only: xref of catalog (root)
Document.pdf_trailer() PDF only: trailer source
Document.prev_location() return (chapter, pno) of preceeding page
Document.reload_page() PDF only: provide a new copy of a page PDF only: save the document
Document.saveIncr() PDF only: save the document incrementally
Document.scrub() PDF only: remove sensitive data
Document.search_page_for() search for a string on a page PDF only: select a subset of pages
Document.set_layer_ui_config() PDF only: set OCG visibility temporarily
Document.set_metadata() PDF only: set the metadata
Document.set_layer() PDF only: mass changing OCG states
Document.set_oc() PDF only: attach OCG/OCMD to image / form xobject
Document.set_ocmd() PDF only: create or update an OCMD
Document.set_page_labels() PDF only: add/update page label definitions
Document.set_toc_item() PDF only: change a single TOC item
Document.set_toc() PDF only: set the table of contents (TOC)
Document.set_xml_metadata() PDF only: create or update document XML metadata
Document.subset_fonts() PDF only: create font subsets
Document.switch_layer() PDF only: activate OC configuration
Document.tobytes() PDF only: writes document to memory
Document.xref_object() PDF only: get the definition source of xref
Document.xref_get_key() PDF only: get the value of a dictionary key
Document.xref_get_keys() PDF only: list the keys of object at xref
Document.xref_set_key() PDF only: set the value of a dictionary key
Document.xref_stream_raw() PDF only: raw stream source at xref
Document.chapter_count number of chapters
Document.FormFonts PDF only: list of global widget fonts
Document.is_closed has document been closed?
Document.is_dirty PDF only: has document been changed yet?
Document.is_encrypted document (still) encrypted?
Document.is_form_pdf is this a Form PDF?
Document.is_pdf is this a PDF?
Document.is_reflowable is this a reflowable document?
Document.is_repaired PDF only: has this PDF been repaired during open?
Document.last_location (chapter, pno) of last page
Document.metadata metadata filename of document
Document.needs_pass require password to access data?
Document.outline first Outline item
Document.page_count number of pages
Document.permissions permissions to access the document

Class API

class Document
__init__(self, filename=None, stream=None, filetype=None, rect=None, width=0, height=0, fontsize=11)

Creates a Document object.

  • With default parameters, a new empty PDF document will be created.
  • If stream is given, then the document is created from memory and either filename or filetype must indicate its type.
  • If stream is None, then a document is created from the file given by filename. Its type is inferred from the extension, which can be overruled by specifying filetype.
  • filename (str,pathlib) – A UTF-8 string or pathlib object containing a file path (or a file type, see below).
  • stream (bytes,bytearray,BytesIO) –

    A memory area containing a supported document. Its type must be specified by either filename or filetype.

    (Changed in version 1.14.13) io.BytesIO is now also supported.

  • filetype (str) – A string specifying the type of document. This may be something looking like a filename (e.g. “x.pdf”), in which case MuPDF uses the extension to determine the type, or a mime type like application/pdf. Just using strings like “pdf” will also work.
  • rect (rect_like) – a rectangle specifying the desired page size. This parameter is only meaningful for documents with a variable page layout (“reflowable” documents), like e-books or HTML, and ignored otherwise. If specified, it must be a non-empty, finite rectangle with top-left coordinates (0, 0). Together with parameter fontsize, each page will be accordingly laid out and hence also determine the number of pages.
  • width (float) – may used together with height as an alternative to rect to specify layout information.
  • height (float) – may used together with width as an alternative to rect to specify layout information.
  • fontsize (float) – the default fontsize for reflowable document types. This parameter is ignored if none of the parameters rect or width and height are specified. Will be used to calculate the page layout.

Overview of possible forms (open is a synonym of Document):

>>> # from a file
>>> doc ="some.pdf")
>>> doc ="some.file", None, "pdf")  # copes with wrong extension
>>> doc ="some.file", filetype="pdf")  # copes with wrong extension
>>> # from memory
>>> doc ="pdf", mem_area)
>>> doc =, mem_area, "pdf")
>>> doc =, filetype="pdf")
>>> # new empty PDF
>>> doc =

The Document class can be also be used as a context manager. On exit, the document will automatically be closed.

>>> import fitz
>>> with as doc:
        for page in doc: print("page %i" % page.number)
page 0
page 1
page 2
page 3
>>> doc.is_closed

(New in v1.18.4)

Return the cross reference number of an OCG or OCMD attached to an image or form xobject.

Parameters:xref (int) – the xref of an image or form xobject. Valid such cross reference numbers are returned by Document.get_page_images(), resp. Document.get_page_xobjects(). For invalid numbers, an exception is raised.
Return type:int
Returns:the cross reference number of an optional contents object or zero if there is none.
set_oc(xref, ocxref)

(New in v1.18.4)

If xref represents an image or form xobject, set or remove the cross reference number ocxref of an optional contents object.

  • xref (int) – the xref of an image or form xobject [5]. Valid such cross reference numbers are returned by Document.get_page_images(), resp. Document.get_page_xobjects(). For invalid numbers, an exception is raised.
  • ocxref (int) – the xref number of an OCG / OCMD. If not zero, an invalid reference raises an exception. If zero, any OC reference is removed.

(New in v1.18.3)

Show optional layer configurations. There always is a standard one, which is not included in the response.

>>> for item in doc.get_layers(): print(item)
{'number': 0, 'name': 'my-config', 'creator': ''}
>>> # use 'number' as config identifyer in add_ocg
add_layer(name, creator=None, on=None)

(New in v1.18.3)

Add an optional content configuration. Layers serve as a collection of ON / OFF states for optional content groups and allow fast visibility switches between different views on the same document.

  • name (str) – arbitrary name.
  • creator (str) – (optional) creating software.
  • on (sequ) – a sequence of OCG xref numbers which should be set to ON when this layer gets activated. All OCGs not listed here will be set to OFF.
switch_layer(number, as_default=False)

(New in v1.18.3)

Switch to a document view as defined by the optional layer’s configuration number. This is temporary, except if established as default.

  • number (int) – config number as returned by Document.layer_configs().
  • as_default (bool) – make this the default configuration.

Activates the ON / OFF states of OCGs as defined in the identified layer. If as_default=True, then additionally all layers, including the standard one, are merged and the result is written back to the standard layer, and all optional layers are deleted.

add_ocg(name, config=-1, on=True, intent="View", usage="Artwork")

(New in v1.18.3)

Add an optional content group. An OCG is the most important unit of information to determine object visibility. For a PDF, in order to be regarded as having optional content, at least one OCG must exist.

  • name (str) – arbitrary name. Will show up in supporting PDF viewers.
  • config (int) – layer configuration number. Default -1 is the standard configuration.
  • on (bool) – standard visibility status for objects pointing to this OCG.
  • intent (str,list) – a string or list of strings declaring the visibility intents. There are two PDF standard values to choose from: “View” and “Design”. Default is “View”. Correct spelling is important.
  • usage (str) – another influencer for OCG visibility. This will become part of the OCG’s /Usage key. There are two PDF standard values to choose from: “Artwork” and “Technical”. Default is “Artwork”. Please only change when required.

xref of the created OCG. Use as entry for oc parameter in supporting objects.


Multiple OCGs with identical parameters may be created. This will not cause problems. Garbage option 3 of will get rid of any duplicates.

set_ocmd(xref=0, ocgs=None, policy="AnyOn", ve=None)

(New in v1.18.4)

Create or update an OCMD, Optional Content Membership Dictionary.

  • xref (int) – xref of the OCMD to be updated, or 0 for a new OCMD.
  • ocgs (list) – a sequence of xref numbers of existing OCG PDF objects.
  • policy (str) – one of “AnyOn” (default), “AnyOff”, “AllOn”, “AllOff” (mixed or lower case).
  • ve (list) – a “visibility expression”. This is a list of arbitrarily nested other lists – see explanation below. Use as an alternative to the combination ocgs / policy if you need to formulate more complex conditions.
Return type:



xref of the OCMD. Use as oc=xref parameter in supporting objects, and respectively in Document.set_oc() or Annot.set_oc().


Like an OCG, an OCMD has a visibility state ON or OFF, and it can be used like an OCG. In contrast to an OCG, the OCMD state is determined by evaluating the state of one or more OCGs via special forms of boolean expressions. If the expression evaluates to true, the OCMD state is ON and OFF for false.

There are two ways to formulate OCMD visibility:

  1. Use the combination of ocgs and policy: The policy value is interpreted as follows:
  • AnyOn – (default) true if at least one OCG is ON.
  • AnyOff – true if at least one OCG is OFF.
  • AllOn – true if all OCGs are ON.
  • AllOff – true if all OCGs are OFF.

Suppose you want two PDF objects be displayed exactly one at a time (if one is ON, then the other one must be OFF):

Solution: use an OCG for object 1 and an OCMD for object 2. Create the OCMD via set_ocmd(ocgs=[xref], policy="AllOff"), with the xref of the OCG.

  1. Use the visibility expression ve: This is a list of two or more items. The first item is a logical keyword: one of the strings “and”, “or”, or “not”. The second and all subsequent items must either be an integer or another list. An integer must be the xref number of an OCG. A list must again have at least two items starting with one of the boolean keywords. This syntax is a bit awkward, but quite powerful:
  • Each list must start with a logical keyword.
  • If the keyword is a “not”, then the list must have exactly two items. If it is “and” or “or”, any number of other items may follow.
  • Items following the logical keyword may be either integers or again a list. An integer must be the xref of an OCG. A list must conform to the previous rules.


  • set_ocmd(ve=["or", 4, ["not", 5], ["and", 6, 7]]). This delivers ON if the following is true: “4 is ON, or 5 is OFF, or 6 and 7 are both ON”.
  • set_ocmd(ve=["not", xref]). This has the same effect as the OCMD example created under 1.

For more details and examples see page 224 of Adobe PDF References. Also do have a look at example scripts here.

Visibility expressions, /VE, are part of PDF specification version 1.6. So not all PDF viewers / readers may already support this feature and hence will react in some standard way for those cases.


(New in v1.18.4)

Retrieve the definition of an OCMD.

Parameters:xref (int) – the xref of the OCMD.
Return type:dict
Returns:a dictionary with the keys xref, ocgs, policy and ve.

(New in v1.18.3)

List of optional content groups by status in the specified configuration. This is a dictionary with lists of cross reference numbers for OCGs that occur in the arrays /ON, /OFF or in some radio button group (/RBGroups).

Parameters:config (int) – the configuration layer (default is the standard config layer).
>>> pprint(doc.get_layer())
{'off': [8, 9, 10], 'on': [5, 6, 7], 'rbgroups': [[7, 10]]}
set_layer(config, on=None, off=None, basestate=None, rbgroups=None)

(New in v1.18.3)

Mass status changes of optional content groups. Permanently sets the status of OCGs.

  • config (int) – desired configuration layer, choose -1 for the default one.
  • on (list) – list of xref of OCGs to set ON. Replaces previous values. An empty list will cause no OCG being set to ON anymore. Should be specified if basestate="ON" is used.
  • off (list) – list of xref of OCGs to set OFF. Replaces previous values. An empty list will cause no OCG being set to OFF anymore. Should be specified if basestate="OFF" is used.
  • basestate (str) – desired state of OCGs that are not mentioned in on resp. off. Possible values are “ON”, “OFF” or “Unchanged”. Upper / lower case possible.
  • rbgroups (list) – a list of lists. Replaces previous values. Each sublist should contain two or more OCG xrefs. OCGs in the same sublist are handled like buttons in a radio button group: setting one to ON automatically sets all other group members to OFF.

Values None will not change the corresponding PDF array.

>>> doc.set_layer(-1, basestate="OFF")  # only changes the base state
>>> pprint(doc.get_layer())
{'basestate': 'OFF', 'off': [8, 9, 10], 'on': [5, 6, 7], 'rbgroups': [[7, 10]]}

(New in v1.18.3)

Details of all optional content groups. This is a dictionary of dictionaries like this (key is the OCG’s xref):

>>> pprint(doc.get_ocgs())
{13: {'on': True,
      'intent': ['View', 'Design'],
      'name': 'Circle',
      'usage': 'Artwork'},
14: {'on': True,
      'intent': ['View', 'Design'],
      'name': 'Square',
      'usage': 'Artwork'},
15: {'on': False, 'intent': ['View'], 'name': 'Square', 'usage': 'Artwork'}}

(New in v1.18.3)

Show the visibility status of optional content that is modifyable by the user interface of supporting PDF viewers. Example:

>>> pprint(doc.layer_ui_configs())
 ({'depth': 0,
  'locked': False,
  'number': 0,
  'on': True,
  'text': 'Circle',
  'type': 'checkbox'},
 {'depth': 0,
  'locked': False,
  'number': 1,
  'on': False,
  'text': 'Square',
  'type': 'checkbox'})
 >>> # refers to OCGs named "Circle" (ON), resp. "Square" (OFF)


  • Only reports items contained in the currently selected layer configuration.
  • The meaning of the dictionary keys is as follows:
    • depth: item’s nesting level in the /Order array
    • locked: whether changing the item’s state is prohibited
    • number: running sequence number
    • on: item state
    • text: text string or name field of the originating OCG
    • type: one of “label” (set by a text string), “checkbox” (set by a single OCG) or “radiobox” (set by a set of connected OCGs)
set_layer_ui_config(number, action=0)

(New in v1.18.3)

Modify OC visibility status of content groups. This is analog to what supporting PDF viewers would offer.


Visibility is not a property stored with the OCG. It is not even an information necessarily present in the PDF document at all. Instead, the current visibility is temporarily set using the user interface of some supporting PDF consumer software. The same type of functionality is offered by this method.

To make permanent changes, use Document.set_layer().

  • number (in) – number as returned by Document.layer_ui_configs().
  • action (int) – 0 = set on (default), 1 = toggle on/off, 2 = set off.


>>> # let's make above "Square" visible:
>>> doc.set_layer_ui_config(1, action=0)
>>> pprint(doc.layer_ui_configs())
({'depth': 0,
  'locked': False,
  'number': 0,
  'on': True,
  'text': 'Circle',
  'type': 'checkbox'},
{'depth': 0,
  'locked': False,
  'number': 1,
  'on': True,  # <===
  'text': 'Square',
  'type': 'checkbox'})

Decrypts the document with the string password. If successful, document data can be accessed. For PDF documents, the “owner” and the “user” have different priviledges, and hence different passwords may exist for these authorization levels. The method will automatically establish the appropriate (owner or user) access rights for the provided password.

Parameters:password (str) – owner or user password.
Return type:int
Returns:a positive value if successful, zero otherwise (the string does not match either password). If positive, the indicator Document.is_encrypted is set to False. Positive return codes carry the following information detail:
  • 1 => authenticated, but the PDF has neither owner nor user passwords.
  • 2 => authenticated with the user password.
  • 4 => authenticated with the owner password.
  • 6 => authenticated and both passwords are equal – probably a rare situation.


The document may be protected by an owner, but not by a user password. Detect this situation via doc.authenticate(“”) == 2. This allows opening and reading the document without authentication, but, depending on the Document.permissions value, other actions may be prohibited. PyMuPDF (like MuPDF) in this case ignores those restrictions. So, – in contrast to any PDF viewers – you can for example extract text and add or modify content, even if the respective permission flags PDF_PERM_COPY, PDF_PERM_MODIFY, PDF_PERM_ANNOTATE, etc. are set off! It is your responsibility building a legally compliant application where applicable.

get_page_numbers(label, only_one=False)

(New in v 1.18.6)

PDF only: Return a list of page numbers that have the specified label – note that labels may not be unique in a PDF. This implies a sequential search through all page numbers to compare their labels.


Implementation detail – pages are not loaded for this purpose.

  • label (str) – the label to look for, e.g. “vii” (Roman number 7).
  • only_one (bool) – stop after first hit. Useful e.g. if labelling is known to be unique, or there are many pages, etc. The default will check every page number.
Return type:



list of page numbers that have this label. Empty if none found, no labels defined, etc.


(New in v1.18.7)

PDF only: Extract the list of page label definitions. Typically used for modifications before feeding it into Document.set_page_labels().

Returns:a list of dictionaries as defined in Document.set_page_labels().

(New in v1.18.6)

PDF only: Add or update the page label definitions of the PDF.

Parameters:labels (list) –

a list of dictionaries. Each dictionary defines a label building rule and a 0-based “start” page number. That start page is the first for which the label definition is valid. Each dictionary has up to 4 items and looks like {'startpage': int, 'prefix': str, 'style': str, 'firstpagenum': int} and has the following items.

  • startpage: (int) the first page number (0-based) to apply the label rule. This key must be present. The rule is applied to all subsequent pages until either end of document or superseded by the rule with the next larger page number.
  • prefix: (str) an arbitrary string to start the label with, e.g. “A-“. Default is “”.
  • style: (str) the numbering style. Available are “D” (decimal), “r”/”R” (Roman numbers, lower / upper case), and “a”/”A” (lower / upper case alphabetical numbering: “a” through “z”, then “aa” through “az”, etc.). Default is “”. If “”, no numbering will take place and the pages in that range will receive the same label consisting of the prefix value. If prefix is also omitted, then the label will be “”.
  • firstpagenum: (int) start numbering with this value. Default is 1, smaller values are ignored.

For example:

[{'startpage': 6, 'prefix': 'A-', 'style': 'D', 'firstpagenum': 10},
 {'startpage': 10, 'prefix': '', 'style': 'D', 'firstpagenum': 1}]

will generate the labels “A-10”, “A-11”, “A-12”, “A-13”, “1”, “2”, “3”, … for pages 6, 7 and so on until end of document. Pages 0 through 5 will have the label “”.


(New in v.1.17.3) Return a page pointer in a reflowable document. After re-layouting the document, the result of this method can be used to find the new location of the page.


Do not confuse with items of a table of contents, TOC.

Parameters:loc (list,tuple) – page location. Must be a valid (chapter, pno).
Return type:pointer
Returns:a long integer in pointer format. To be used for finding the new location of the page after re-layouting the document. Do not touch or re-assign.

(New in v.1.17.3) Return the new page location after re-layouting the document.

Parameters:bookmark (pointer) – created by Document.make_bookmark().
Return type:tuple
Returns:the new (chapter, pno) of the page.

(New in v.1.17.0) Return the number of pages of a chapter.

Parameters:chapter (int) – the 0-based chapter number.
Return type:int
Returns:number of pages in chapter. Relevant only for document types whith chapter support (EPUB currently).

(New in v.1.17.0) Return the location of the following page.

Parameters:page_id (tuple) – the current page id. This must be a tuple (chapter, pno) identifying an existing page.
Returns:The tuple of the following page, i.e. either (chapter, pno + 1) or (chapter + 1, 0), or the empty tuple () if the argument was the last page. Relevant only for document types whith chapter support (EPUB currently).

(New in v.1.17.0) Return the locator of the preceeding page.

Parameters:page_id (tuple) – the current page id. This must be a tuple (chapter, pno) identifying an existing page.
Returns:The tuple of the preceeding page, i.e. either (chapter, pno - 1) or the last page of the receeding chapter, or the empty tuple () if the argument was the first page. Relevant only for document types whith chapter support (EPUB currently).

Create a Page object for further processing (like rendering, text searching, etc.).

(Changed in v1.17.0) For document types supporting a so-called “chapter structure” (like EPUB), pages can also be loaded via the combination of chapter number and relative page number, instead of the absolute page number. This should significantly speed up access for large documents.

Parameters:page_id (int,tuple) –

(Changed in v1.17.0)

Either a 0-based page number, or a tuple (chapter, pno). For an integer, any -∞ < page_id < page_count is acceptable. While page_id is negative, page_count will be added to it. For example: to load the last page, you can use doc.load_page(-1). After this you have page.number = doc.page_count - 1.

For a tuple, chapter must be in range Document.chapter_count, and pno must be in range Document.chapter_page_count() of that chapter. Both values are 0-based. Using this notation, Page.number will equal the given tuple. Relevant only for document types whith chapter support (EPUB currently).

Return type:Page


Documents also follow the Python sequence protocol with page numbers as indices: doc.load_page(n) == doc[n].

For absolute page numbers only, expressions like “for page in doc: …” and “for page in reversed(doc): …” will successively yield the document’s pages. Refer to Document.pages() which allows processing pages as with slicing.

You can also use index notation with the new chapter-based page identification: use page = doc[(5, 2)] to load the third page of the sixth chapter.

To maintain a consistent API, for document types not supporting a chapter structure (like PDFs), Document.chapter_count is 1, and pages can also be loaded via tuples (0, pno). See this [3] footnote for comments on performance improvements.


(New in version 1.16.10)

PDF only: Provide a new copy of a page after finishing and updating all pending changes.

Parameters:page (Page) – page object.
Return type:Page
Returns:a new copy of the same page. All pending updates (e.g. to annotations or widgets) will be finalized and a fresh copy of the page will be loaded.


In a typical use case, a page Pixmap should be taken after annotations / widgets have been added or changed. To force all those changes being reflected in the page structure, this method re-instates a fresh copy while keeping the object hierarchy “document -> page -> annotations/widgets” intact.


(New in version 1.17.7)

PDF only: Return the unrotated page rectangle – without loading the page (via Document.load_page()). This is meant for internal purpose requiring best possible performance.

Parameters:pno (int) – 0-based page number.
Returns:Rect of the page like Page.rect(), but ignoring any rotation.

(New in version 1.17.7)

PDF only: Return the xref of the page – without loading the page (via Document.load_page()). This is meant for internal purpose requiring best possible performance.

Parameters:pno (int) – 0-based page number.
Returns:xref of the page like Page.xref.
pages(start=None[, stop=None[, step=None]])

(New in version 1.16.4)

A generator for a range of pages. Parameters have the same meaning as in the built-in function range(). Intended for expressions of the form “for page in doc.pages(start, stop, step): …”.

  • start (int) – start iteration with this page number. Default is zero, allowed values are -∞ < start < page_count. While this is negative, page_count is added before starting the iteration.
  • stop (int) – stop iteration at this page number. Default is page_count, possible are -∞ < stop <= page_count. Larger values are silently replaced by the default. Negative values will cyclically emit the pages in reversed order. As with the built-in range(), this is the first page not returned.
  • step (int) – stepping value. Defaults are 1 if start < stop and -1 if start > stop. Zero is not allowed.

a generator iterator over the document’s pages. Some examples:

  • ”doc.pages()” emits all pages.
  • ”doc.pages(4, 9, 2)” emits pages 4, 6, 8.
  • ”doc.pages(0, None, 2)” emits all pages with even numbers.
  • ”doc.pages(-2)” emits the last two pages.
  • ”doc.pages(-1, -1)” emits all pages in reversed order.
  • ”doc.pages(-1, -10)” always emits 10 pages in reversed order, starting with the last page – repeatedly if the document has less than 10 pages. So for a 4-page document the following page numbers are emitted: 3, 2, 1, 0, 3, 2, 1, 0, 3, 2, 1, 0, 3.

convert_to_pdf(from_page=-1, to_page=-1, rotate=0)

Create a PDF version of the current document and write it to memory. All document types are supported. The parameters have the same meaning as in insert_pdf(). In essence, you can restrict the conversion to a page subset, specify page rotation, and revert page sequence.

  • from_page (int) – first page to copy (0-based). Default is first page.
  • to_page (int) – last page to copy (0-based). Default is last page.
  • rotate (int) – rotation angle. Default is 0 (no rotation). Should be n * 90 with an integer n (not checked).
Return type:



a Python bytes object containing a PDF file image. It is created by internally using tobytes(garbage=4, deflate=True). See tobytes(). You can output it directly to disk or open it as a PDF. Here are some examples:

>>> # convert an XPS file to PDF
>>> xps ="some.xps")
>>> pdfbytes = xps.convert_to_pdf()
>>> # either do this -->
>>> pdf ="pdf", pdfbytes)
>>> # or this -->
>>> pdfout = open("some.pdf", "wb")
>>> pdfout.tobytes(pdfbytes)
>>> pdfout.close()
>>> # copy image files to PDF pages
>>> # each page will have image dimensions
>>> doc =                     # new PDF
>>> imglist = [ ... image file names ...] # e.g. a directory listing
>>> for img in imglist:           # open image as a document
        pdfbytes=imgdoc.convert_to_pdf()  # make a 1-page PDF of it"pdf", pdfbytes)
        doc.insert_pdf(imgpdf)             # insert the image PDF


The method uses the same logic as the mutool convert CLI. This works very well in most cases – however, beware of the following limitations.

  • Image files: perfect, no issues detected. Apparently however, image transparency is ignored. If you need that (like for a watermark), use Page.insert_image() instead. Otherwise, this method is recommended for its much better prformance.
  • XPS: appearance very good. Links work fine, outlines (bookmarks) are lost, but can easily be recovered [2].
  • EPUB, CBZ, FB2: similar to XPS.
  • SVG: medium. Roughly comparable to svglib.

Creates a table of contents (TOC) out of the document’s outline chain.

Parameters:simple (bool) – Indicates whether a simple or a detailed TOC is required. If False, each item of the list also contains a dictionary with linkDest details for each outline entry.
Return type:list
Returns:a list of lists. Each entry has the form [lvl, title, page, dest]. Its entries have the following meanings:
  • lvl – hierarchy level (positive int). The first entry is always 1. Entries in a row are either equal, increase by 1, or decrease by any number.
  • title – title (str)
  • page – 1-based page number (int). If -1 either no destination or outside document.
  • dest – (dict) included only if simple=False. Contains details of the TOC item as follows:
    • kind: destination kind, see Link Destination Kinds.
    • file: filename if kind is LINK_GOTOR or LINK_LAUNCH.
    • page: target page, 0-based, LINK_GOTOR or LINK_GOTO only.
    • to: position on target page (Point).
    • zoom: (float) zoom factor on target page.
    • xref: xref of the item (0 if no PDF).
    • color: item color in PDF RGB format (red, green, blue), or omitted (always omitted if no PDF).
    • bold: true if bold item text or omitted. PDF only.
    • italic: true if italic item text, or omitted. PDF only.
    • collapse: true if sub-items are folded, or omitted. PDF only.

(New in v1.18.7)

PDF only: Return the PDF dictionary keys of the object provided by its xref number.

Parameters:xref (int) – the xref. (Changed in v1.18.10) Use -1 to access the special dictionary “PDF trailer”.
Returns:a tuple of dictionary keys present in object xref. Examples:
>>> from pprint import pprint
>>> import fitz
>>> xref = doc.page_xref(0)  # xref of page 0
>>> pprint(doc.xref_get_keys(xref))  # primary level keys of a page
('Type', 'Contents', 'Resources', 'MediaBox', 'Parent')
>>> pprint(doc.xref_get_keys(-1))  # primary level keys of the trailer
('Type', 'Index', 'Size', 'W', 'Root', 'Info', 'ID', 'Length', 'Filter')
xref_get_key(xref, key)

(New in v1.18.7)

PDF only: Return type and value of a PDF dictionary key of an xref.

  • xref (int) – the xref. Changed in v1.18.10: Use -1 to access the special dictionary “PDF trailer”.
  • key (str) – the desired PDF key. Must exactly match (case-sensitive) one of the keys contained in Document.xref_get_keys().

a tuple (type, value) of strings, where type is one of “xref”, “array”, “dict”, “int”, “float”, “null”, “bool”, “name”, “string” or “unknown” (should not occur). Independent of “type”, the value of the key is always formatted as a string – see the following example – and (almost always) a faithful reflection of what is stored in the PDF. In most cases, the format of the value string also gives a clue about the key type:

  • A “name” always starts with a “/” slash.

  • An “xref” always ends with ” 0 R”.

  • An “array” is always enclosed in “[…]” brackets.

  • A “dict” is always enclosed in “<<…>>” brackets.

  • A “bool”, resp. “null” always equal either “true”, “false”, resp. “null”.

  • ”float” and “int” are represented by their string format – and are thus not always distinguishable.

  • A “string” is converted to UTF-8 and may therefore deviate from what is stored in the PDF. For example, the PDF key “Author” may have a value of “<FEFF004A006F0072006A00200058002E0020004D0063004B00690065>” in the file, but the method will return ('string', 'Jorj X. McKie').

    >>> for key in doc.xref_get_keys(xref):
            print(key, "=" , doc.xref_get_key(xref, key))
    Type = ('name', '/Page')
    Contents = ('xref', '1297 0 R')
    Resources = ('xref', '1296 0 R')
    MediaBox = ('array', '[0 0 612 792]')
    Parent = ('xref', '1301 0 R')
    >>> #
    >>> # Now same thing for the PDF trailer.
    >>> # It has no xref, so -1 must be used instead.
    >>> #
    >>> for key in doc.xref_get_keys(-1):
            print(key, "=", doc.xref_get_key(-1, key))
    Type = ('name', '/XRef')
    Index = ('array', '[0 8802]')
    Size = ('int', '8802')
    W = ('array', '[1 3 1]')
    Root = ('xref', '8799 0 R')
    Info = ('xref', '8800 0 R')
    ID = ('array', '[<DC9D56A6277EFFD82084E64F9441E18C><DC9D56A6277EFFD82084E64F9441E18C>]')
    Length = ('int', '21111')
    Filter = ('name', '/FlateDecode')

xref_set_key(xref, key, value)
  • New in v1.18.7, changed in v 1.18.13
  • Changed in v1.19.4: remove a key “physically” if set to “null”.

PDF only: Set (add, update, delete) the value of a PDF key for the object given by an xref.


This is an expert function: if you do not know what you are doing, there is a high risk to render (parts of) the PDF unusable. Please do consult Adobe PDF References about object specification formats (page 18) and the structure of special dictionary types like page objects.

  • xref (int) – the xref. Changed in v1.18.13: To update the PDF trailer, specify -1.
  • key (str) – the desired PDF key (without leading “/”). Must not be empty. Any valid PDF key – whether already present in the object (which will be overwritten) – or new. It is possible to use PDF path notation like "Resources/ExtGState" – which sets the value for key "/ExtGState" as a sub-object of "/Resources".
  • value (str) –

    the value for the key. It must be a non-empty string and, depending on the desired PDF object type, the following rules must be observed. There is some syntax checking, but no type checking and no checking if it makes sense PDF-wise, i.e. no semantics checking. Upper / lower case is important!

    • xref – must be provided as "nnn 0 R" with a valid xref number nnn of the PDF. The suffix “0 R” is required to be recognizable as an xref by PDF applications.
    • array – a string like "[a b c d e f]". The brackets are required. Array items must be separated by at least one space (not commas like in Python). An empty array "[]" is possible and equivalent to removing the key. Array items may be any PDF objects, like dictionaries, xrefs, other arrays, etc. Like in Python, array items may be of different types.
    • dict – a string like "<< ... >>". The brackets are required and must enclose a valid PDF dictionary definition. The empty dictionary "<<>>" is possible and equivalent to removing the key.
    • int – an integer formatted as a string.
    • float – a float formatted as a string. Scientific notation (with exponents) is not allowed by PDF.
    • null – the string "null". This is the PDF equivalent to Python’s None and causes the key to be ignored – however not necessarily removed, resp. removed on saves with garbage collection. Changed in v1.19.4: If the key is no path hierarchy (i.e. contains no slash “/”), then it will be completely removed.
    • bool – one of the strings "true" or "false".
    • name – a valid PDF name with a leading slash: "/PageLayout". See page 16 of the Adobe PDF References.
    • string – a valid PDF string. All PDF strings must be enclosed by brackets. Denote the empty string as "()". Depending on its content, the possible brackets are
      • ”(…)” for ASCII-only text. Reserved PDF characters must be backslash-escaped and non-ASCII characters must be provided as 3-digit backslash-escaped octals – including leading zeros. Example: 12 = 0x0C must be encoded as \014.
      • ”<…>” for hex-encoded text. Every character must be represented by two hex-digits (lower or upper case).
      • If in doubt, we strongly recommend to use get_pdf_str()! This function automatically generates the right brackets, escapes, and overall format. It will for example do conversions like these:
        >>> # because of the € symbol, the following yields UTF-16BE BOM
        >>> fitz.get_pdf_str("Pay in $ or €.")
        >>> # escapes for brackets and non-ASCII
        >>> fitz.get_pdf_str("Prices in EUR (USD also accepted). Areas are in m².")
        '(Prices in EUR \\(USD also accepted\\). Areas are in m\\262.)'
get_page_pixmap(pno, *args, **kwargs)

Creates a pixmap from page pno (zero-based). Invokes Page.get_pixmap().

Parameters:pno (int) – page number, 0-based in -∞ < pno < page_count.
Return type:Pixmap

(Changed in v1.18.11)

PDF only: (New in v1.16.13) Return a list of all XObjects referenced by a page.

Parameters:pno (int) – page number, 0-based, -∞ < pno < page_count.
Return type:list
Returns:a list of (non-image) XObjects. These objects typically represent pages embedded (not copied) from other PDFs. For example, Page.show_pdf_page() will create this type of object. An item of this list has the following layout: (xref, name, invoker, bbox), where
  • xref (int) is the XObject’s xref.
  • name (str) is the symbolic name to reference the XObject.
  • invoker (int) the xref of the invoking XObject or zero if the page directly invokes it.
  • bbox (Rect) the boundary box of the XObject’s location on the page in untransformed coordinates. To get actual, non-rotated page coordinates, multiply with the page’s transformation matrix Page.transformation_matrix. Changed in v.18.11: the bbox is now formatted as Rect.
get_page_images(pno, full=False)

PDF only: Return a list of all images (directly or indirectly) referenced by the page.

  • pno (int) – page number, 0-based, -∞ < pno < page_count.
  • full (bool) – whether to also include the referencer’s xref (which is zero if this is the page).
Return type:



a list of images referenced by this page. Each item looks like

(xref, smask, width, height, bpc, colorspace, alt. colorspace, name, filter, referencer)


  • xref (int) is the image object number
  • smask (int) is the object number of its soft-mask image
  • width and height (ints) are the image dimensions
  • bpc (int) denotes the number of bits per component (normally 8)
  • colorspace (str) a string naming the colorspace (like DeviceRGB)
  • alt. colorspace (str) is any alternate colorspace depending on the value of colorspace
  • name (str) is the symbolic name by which the image is referenced
  • filter (str) is the decode filter of the image (Adobe PDF References, pp. 22).
  • referencer (int) the xref of the referencer. Zero if directly referenced by the page. Only present if full=True.


In general, this is not the list of images that are actually displayed. This method only parses several PDF objects to collect references to embedded images. It does not analyse the page’s contents, where all the actual image display commands are defined. To get this information, please use Page.get_image_info(). Also have a look at the discussion in section Structure of Dictionary Outputs.

get_page_fonts(pno, full=False)

PDF only: Return a list of all fonts (directly or indirectly) referenced by the page.

  • pno (int) – page number, 0-based, -∞ < pno < page_count.
  • full (bool) – whether to also include the referencer’s xref. If True, the returned items are one entry longer. Use this option if you need to know, whether the page directly references the font. In this case the last entry is 0. If the font is referenced by an /XObject of the page, you will find its xref here.
Return type:



a list of fonts referenced by this page. Each entry looks like

(xref, ext, type, basefont, name, encoding, referencer),


  • xref (int) is the font object number (may be zero if the PDF uses one of the builtin fonts directly)
  • ext (str) font file extension (e.g. “ttf”, see Font File Extensions)
  • type (str) is the font type (like “Type1” or “TrueType” etc.)
  • basefont (str) is the base font name,
  • name (str) is the symbolic name, by which the font is referenced
  • encoding (str) the font’s character encoding if different from its built-in encoding (Adobe PDF References, p. 254):
  • referencer (int optional) the xref of the referencer. Zero if directly referenced by the page, otherwise the xref of an XObject. Only present if full=True.


>>> pprint(doc.get_page_fonts(0, full=False))
[(12, 'ttf', 'TrueType', 'FNUUTH+Calibri-Bold', 'R8', ''),
 (13, 'ttf', 'TrueType', 'DOKBTG+Calibri', 'R10', ''),
 (14, 'ttf', 'TrueType', 'NOHSJV+Calibri-Light', 'R12', ''),
 (15, 'ttf', 'TrueType', 'NZNDCL+CourierNewPSMT', 'R14', ''),
 (16, 'ttf', 'Type0', 'MNCSJY+SymbolMT', 'R17', 'Identity-H'),
 (17, 'cff', 'Type1', 'UAEUYH+Helvetica', 'R20', 'WinAnsiEncoding'),
 (18, 'ttf', 'Type0', 'ECPLRU+Calibri', 'R23', 'Identity-H'),
 (19, 'ttf', 'Type0', 'TONAYT+CourierNewPSMT', 'R27', 'Identity-H')]


  • This list has no duplicate entries: the combination of xref, name and referencer is unique.
  • In general, this is a superset of the fonts actually in use by this page. The PDF creator may e.g. have specified some global list, of which each page only makes partial use.
get_page_text(pno, output="text", flags=3, textpage=None, sort=False)

Extracts the text of a page given its page number pno (zero-based). Invokes Page.get_text().

Parameters:pno (int) – page number, 0-based, any value -∞ < pno < page_count.

For other parameter refer to the page method.

Return type:str
layout(rect=None, width=0, height=0, fontsize=11)

Re-paginate (“reflow”) the document based on the given page dimension and fontsize. This only affects some document types like e-books and HTML. Ignored if not supported. Supported documents have True in property is_reflowable.

  • rect (rect_like) – desired page size. Must be finite, not empty and start at point (0, 0).
  • width (float) – use it together with height as alternative to rect.
  • height (float) – use it together with width as alternative to rect.
  • fontsize (float) – the desired default fontsize.

PDF only: Keeps only those pages of the document whose numbers occur in the list. Empty sequences or elements outside range(doc.page_count) will cause a ValueError. For more details see remarks at the bottom or this chapter.

Parameters:s (sequence) – The sequence (see Using Python Sequences as Arguments in PyMuPDF) of page numbers (zero-based) to be included. Pages not in the sequence will be deleted (from memory) and become unavailable until the document is reopened. Page numbers can occur multiple times and in any order: the resulting document will reflect the sequence exactly as specified.


  • Page numbers in the sequence need not be unique nor be in any particular order. This makes the method a versatile utility to e.g. select only the even or the odd pages or meeting some other criteria and so forth.
  • On a technical level, the method will always create a new pagetree.
  • When dealing with only a few pages, methods copy_page(), move_page(), delete_page() are easier to use. In fact, they are also much faster – by at least one order of magnitude when the document has many pages.

PDF only: Sets or updates the metadata of the document as specified in m, a Python dictionary.

Parameters:m (dict) – A dictionary with the same keys as metadata (see below). All keys are optional. A PDF’s format and encryption method cannot be set or changed and will be ignored. If any value should not contain data, do not specify its key or set the value to None. If you use {} all metadata information will be cleared to the string “none”. If you want to selectively change only some values, modify a copy of doc.metadata and use it as the argument. Arbitrary unicode values are possible if specified as UTF-8-encoded.

(Changed in v1.18.4) Empty values or “none” are no longer written, but completely omitted.


PDF only: Get the document XML metadata.

Return type:str
Returns:XML metadata of the document. Empty string if not present or not a PDF.

PDF only: Sets or updates XML metadata of the document.

Parameters:xml (str) – the new XML metadata. Should be XML syntax, however no checking is done by this method and any string is accepted.
set_toc(toc, collapse=1)

PDF only: Replaces the complete current outline tree (table of contents) with the one provided as the argument. After successful execution, the new outline tree can be accessed as usual via Document.get_toc() or via Document.outline. Like with other output-oriented methods, changes become permanent only via save() (incremental save supported). Internally, this method consists of the following two steps. For a demonstration see example below.

  • Step 1 deletes all existing bookmarks.
  • Step 2 creates a new TOC from the entries contained in toc.
  • toc (sequence) –

    A list / tuple with all bookmark entries that should form the new table of contents. Output variants of get_toc() are acceptable. To completely remove the table of contents specify an empty sequence or None. Each item must be a list with the following format.

    • [lvl, title, page [, dest]] where
      • lvl is the hierarchy level (int > 0) of the item, which must be 1 for the first item and at most 1 larger than the previous one.
      • title (str) is the title to be displayed. It is assumed to be UTF-8-encoded (relevant for multibyte code points only).
      • page (int) is the target page number (attention: 1-based). Must be in valid range if positive. Set it to -1 if there is no target, or the target is external.
      • dest (optional) is a dictionary or a number. If a number, it will be interpreted as the desired height (in points) this entry should point to on the page. Use a dictionary (like the one given as output by get_toc(False)) for a detailed control of the bookmark’s properties, see Document.get_toc() for a description.
  • collapse (int) – (new in version 1.16.9) controls the hierarchy level beyond which outline entries should initially show up collapsed. The default 1 will hence only display level 1, higher levels must be unfolded using the PDF viewer. To unfold everything, specify either a large integer, 0 or None.
Return type:



the number of inserted, resp. deleted items.


(New in v1.17.7)

PDF only: Return the xref of the outline item. This is mainly used for internal purposes.

arg int idx: index of the item in list Document.get_toc().

  • New in v1.17.7
  • Changed in v1.18.14: no longer remove the item’s text, but show it grayed-out.

PDF only: Remove this TOC item. This is a high-speed method, which disables the respective item, but leaves the overall TOC struture intact. Physically, the item still exists in the TOC tree, but is shown grayed-out and will no longer point to any destination.

This also implies that you can reassign the item to a new destination using Document.set_toc_item(), when required.

Parameters:idx (int) – the index of the item in list Document.get_toc().
set_toc_item(idx, dest_dict=None, kind=None, pno=None, uri=None, title=None, to=None, filename=None, zoom=0)
  • New in v1.17.7
  • Changed in v1.18.6

PDF only: Changes the TOC item identified by its index. Change the item title, destination, appearance (color, bold, italic) or collapsing sub-items – or to remove the item altogether.

Use this method if you need specific changes for selected entries only and want to avoid replacing the complete TOC. This is beneficial especially when dealing with large table of contents.

  • idx (int) – the index of the entry in the list created by Document.get_toc().
  • dest_dict (dict) – the new destination. A dictionary like the last entry of an item in doc.get_toc(False). Using this as a template is recommended. When given, all other parameters are ignored – except title.
  • kind (int) – the link kind, see Link Destination Kinds. If LINK_NONE, then all remaining parameter will be ignored, and the TOC item will be removed – same as Document.del_toc_item(). If None, then only the title is modified and the remaining parameters are ignored. All other values will lead to making a new destination dictionary using the subsequent arguments.
  • pno (int) – the 1-based page number, i.e. a value 1 <= pno <= doc.page_count. Required for LINK_GOTO.
  • uri (str) – the URL text. Required for LINK_URI.
  • title (str) – the desired new title. None if no change.
  • to (point_like) – (optional) points to a coordinate on the arget page. Relevant for LINK_GOTO. If omitted, a point near the page’s top is chosen.
  • filename (str) – required for LINK_GOTOR and LINK_LAUNCH.
  • zoom (float) – use this zoom factor when showing the target page.

Example use: Change the TOC of the SWIG manual to achieve this:

Collapse everything below top level and show the chapter on Python support in red, bold and italic:

>>> import fitz
>>> toc = doc.get_toc(False)  # we need the detailed TOC
>>> # list of level 1 indices and their titles
>>> lvl1 = [(i, item[1]) for i, item in enumerate(toc) if item[0] == 1]
>>> for i, title in lvl1:
        d = toc[i][3]  # get the destination dict
        d["collapse"] = True  # collapse items underneath
        if "Python" in title:  # show the 'Python' chapter
            d["color"] = (1, 0, 0)  # in red,
            d["bold"] = True  # bold and
            d["italic"] = True  # italic
        doc.set_toc_item(i, dest_dict=d)  # update this toc item

In the previous example, we have changed only 42 of the 1240 TOC items of the file.


(New in version 1.16.0)

Check whether the document can be saved incrementally. Use it to choose the right option without encountering exceptions.

scrub(attached_files=True, clean_pages=True, embedded_files=True, hidden_text=True, javascript=True, metadata=True, redactions=True, redact_images=0, remove_links=True, reset_fields=True, reset_responses=True, thumbnails=True, xml_metadata=True)

PDF only: (New in v1.16.14) Remove potentially sensitive data from the PDF. This function is inspired by the similar “Sanitize” function in Adobe Acrobat products. The process is configurable by a number of options, which are all True by default.

  • attached_files (bool) – Search for ‘FileAttachment’ annotations and remove the file content.
  • clean_pages (bool) – Remove any comments from page painting sources. If this option is set to False, then this is also done for hidden_text and redactions.
  • embedded_files (bool) – Remove embedded files.
  • hidden_text (bool) – Remove OCRed text and invisible text [7].
  • javascript (bool) – Remove JavaScript sources.
  • metadata (bool) – Remove PDF standard metadata.
  • redactions (bool) – Apply redaction annotations.
  • redact_images (int) – how to handle images if applying redactions. One of 0 (ignore), 1 (blank out overlaps) or 2 (remove).
  • remove_links (bool) – Remove all links.
  • reset_fields (bool) – Reset all form fields to their defaults.
  • reset_responses (bool) – Remove all responses from all annotations.
  • thumbnails (bool) – Remove thumbnail images from pages.
  • xml_metadata (bool) – Remove XML metadata.
save(outfile, garbage=0, clean=False, deflate=False, deflate_images=False, deflate_fonts=False, incremental=False, ascii=False, expand=0, linear=False, pretty=False, no_new_id=False, encryption=PDF_ENCRYPT_NONE, permissions=-1, owner_pw=None, user_pw=None)
  • Changed in v1.18.7
  • Changed in v1.19.0

PDF only: Saves the document in its current state.

  • outfile (str,Path,fp) – The file path, pathlib.Path or file object to save to. A file object must have been created before via open(...) or io.BytesIO(). Choosing io.BytesIO() is similar to Document.tobytes() below, which equals the getvalue() output of an internally created io.BytesIO().
  • garbage (int) –

    Do garbage collection. Positive values exclude “incremental”.

    • 0 = none
    • 1 = remove unused (unreferenced) objects.
    • 2 = in addition to 1, compact the xref table.
    • 3 = in addition to 2, merge duplicate objects.
    • 4 = in addition to 3, check stream objects for duplication. This may be slow because such data are typically large.
  • clean (bool) – Clean and sanitize content streams [1]. Corresponds to “mutool clean -sc”.
  • deflate (bool) – Deflate (compress) uncompressed streams.
  • deflate_images (bool) – (new in v1.18.3) Deflate (compress) uncompressed image streams [4].
  • deflate_fonts (bool) – (new in v1.18.3) Deflate (compress) uncompressed fontfile streams [4].
  • incremental (bool) – Only save changes to the PDF. Excludes “garbage” and “linear”. Can only be used if outfile is a string or a pathlib.Path and equal to Cannot be used for files that are decrypted or repaired and also in some other cases. To be sure, check Document.can_save_incrementally(). If this is false, saving to a new file is required.
  • ascii (bool) – convert binary data to ASCII.
  • expand (int) –

    Decompress objects. Generates versions that can be better read by some other programs and will lead to larger files.

    • 0 = none
    • 1 = images
    • 2 = fonts
    • 255 = all
  • linear (bool) – Save a linearised version of the document. This option creates a file format for improved performance for Internet access. Excludes “incremental”.
  • pretty (bool) – Prettify the document source for better readability. PDF objects will be reformatted to look like the default output of Document.xref_object().
  • no_new_id (bool) – Suppress the update of the file’s /ID field. If the file happen to have no such field at all, also supporess creation of a new one. Default is False, so every save will lead to an updated file iddentification.
  • permissions (int) – (new in version 1.16.0) Set the desired permission levels. See Document Permissions for possible values. Default is granting all.
  • encryption (int) – (new in version 1.16.0) set the desired encryption method. See PDF encryption method codes for possible values.
  • owner_pw (str) – (new in version 1.16.0) set the document’s owner password. (Changed in v1.18.3) If not provided, the user password is taken if provided.
  • user_pw (str) – (new in version 1.16.0) set the document’s user password.


The method does not check, whether a file of that name already exists, will hence not ask for confirmation, and overwrite the file. It is your responsibility as a programmer to handle this.

ez_save(*args, **kwargs)

(New in v1.18.11)

PDF only: The same as but with the changed defaults deflate=True, garbage=3.


PDF only: saves the document incrementally. This is a convenience abbreviation for, incremental=True, encryption=PDF_ENCRYPT_KEEP).

tobytes(garbage=0, clean=False, deflate=False, deflate_images=False, deflate_fonts=False, ascii=False, expand=0, linear=False, pretty=False, no_new_id=False, encryption=PDF_ENCRYPT_NONE, permissions=-1, owner_pw=None, user_pw=None)
  • Changed in v1.18.7
  • Changed in v1.19.0

PDF only: Writes the current content of the document to a bytes object instead of to a file. Obviously, you should be wary about memory requirements. The meanings of the parameters exactly equal those in save(). Chapter Collection of Recipes contains an example for using this method as a pre-processor to pdfrw.

(Changed in version 1.16.0) for extended encryption support.

Return type:bytes
Returns:a bytes object containing the complete document.
search_page_for(pno, text, quads=False)

Search for “text” on page number “pno”. Works exactly like the corresponding Page.search_for(). Any integer -∞ < pno < page_count is acceptable.

insert_pdf(docsrc, from_page=-1, to_page=-1, start_at=-1, rotate=-1, links=True, annots=True, show_progress=0, final=1)
  • Changed in v1.19.3 - as a fix to issue #537, form fields are always excluded.

PDF only: Copy the page range [from_page, to_page] (including both) of PDF document docsrc into the current one. Inserts will start with page number start_at. Value -1 indicates default values. All pages thus copied will be rotated as specified. Links and annotations can be excluded in the target, see below. All page numbers are 0-based.

  • docsrc (Document) – An opened PDF Document which must not be the current document. However, it may refer to the same underlying file.
  • from_page (int) – First page number in docsrc. Default is zero.
  • to_page (int) – Last page number in docsrc to copy. Defaults to last page.
  • start_at (int) – First copied page, will become page number start_at in the target. Default -1 appends the page range to the end. If zero, the page range will be inserted before current first page.
  • rotate (int) – All copied pages will be rotated by the provided value (degrees, integer multiple of 90).
  • links (bool) – Choose whether (internal and external) links should be included in the copy. Default is True. Internal links to outside the copied page range are always excluded.
  • annots (bool) – (new in version 1.16.1) choose whether annotations should be included in the copy. (Fixed in v1.19.3) Form fields can never be copied.
  • show_progress (int) – (new in version 1.17.7) specify an interval size greater zero to see progress messages on sys.stdout. After each interval, a message like Inserted 30 of 47 pages. will be printed.
  • final (int) – (new in v1.18.0) controls whether the list of already copied objects should be dropped after this method, default True. Set it to 0 except for the last one of multiple insertions from the same source PDF. This saves target file size and speeds up execution considerably.


  1. If from_page > to_page, pages will be copied in reverse order. If 0 <= from_page == to_page, then one page will be copied.
  2. docsrc TOC entries will not be copied. It is easy however, to recover a table of contents for the resulting document. Look at the examples below and at program in the examples directory: it can join PDF documents and at the same time piece together respective parts of the tables of contents.
new_page(pno=-1, width=595, height=842)

PDF only: Insert an empty page.

  • pno (int) – page number in front of which the new page should be inserted. Must be in 1 < pno <= page_count. Special values -1 and doc.page_count insert after the last page.
  • width (float) – page width.
  • height (float) – page height.
Return type:



the created page object.

insert_page(pno, text=None, fontsize=11, width=595, height=842, fontname="helv", fontfile=None, color=None)

PDF only: Insert a new page and insert some text. Convenience function which combines Document.new_page() and (parts of) Page.insert_text().

Parameters:pno (int) –

page number (0-based) in front of which to insert. Must be in range(-1, doc.page_count + 1). Special values -1 and doc.page_count insert after the last page.

Changed in version 1.14.12
This is now a positional parameter

For the other parameters, please consult the aforementioned methods.

Return type:int
Returns:the result of Page.insert_text() (number of successfully inserted lines).

PDF only: Delete a page given by its 0-based number in -∞ < pno < page_count - 1.

  • Changed in v1.18.14: support Python’s del statement.
Parameters:pno (int) – the page to be deleted. Negative number count backwards from the end of the document (like with indices). Default is the last page.
delete_pages(*args, **kwds)
  • Changed in v1.18.13: more flexibility specifying pages to delete.
  • Changed in v1.18.14: support Python’s del statement.

PDF only: Delete multiple pages given as 0-based numbers.

Format 1: Use keywords. Represents the old format. A contiguous range of pages is removed.
  • “from_page”: first page to delete. Zero if omitted.
  • “to_page”: last page to delete. Last page in document if omitted. Must not be less then “from_page”.

Format 2: Two page numbers as positional parameters. Handled like Format 1.

Format 3: One positional integer parameter. Equivalent to Page.delete_page().

Format 4: One positional parameter of type list, tuple or range() of page numbers. The items of this sequence may be in any order and may contain duplicates.

Format 5: (New in v1.18.14) Using the Python del statement and index / slice notation is now possible.


(Changed in v1.14.17, optimized in v1.17.7) In an effort to maintain a valid PDF structure, this method and delete_page() will also deactivate items in the table of contents which point to deleted pages. “Deactivation” here means, that the bookmark will point to nowhere and the title will be shown grayed-out by supporting PDF viewers. The overall TOC structure is left intact.

It will also remove any links on remaining pages which point to a deleted one. This action may have an extended response time for documents with many pages.

Following examples will all delete pages 500 through 519:

  • doc.delete_pages(500, 519)
  • doc.delete_pages(from_page=500, to_page=519)
  • doc.delete_pages((500, 501, 502, ... , 519))
  • doc.delete_pages(range(500, 520))
  • del doc[500:520]
  • del doc[(500, 501, 502, ... , 519)]
  • del doc[range(500, 520)]

For the Adobe PDF References the above takes about 0.6 seconds, because the remaining 1290 pages must be cleaned from invalid links.

In general, the performance of this method is dependent on the number of remaining pages – not on the number of deleted pages: in the above example, deleting all pages except those 20, will need much less time.

copy_page(pno, to=-1)

PDF only: Copy a page reference within the document.

  • pno (int) – the page to be copied. Must be in range 0 <= pno < len(doc).
  • to (int) – the page number in front of which to copy. The default inserts after the last page.


Only a new reference to the page object will be created – not a new page object, all copied pages will have identical attribute values, including the Page.xref. This implies that any changes to one of these copies will appear on all of them.

fullcopy_page(pno, to=-1)

(New in version 1.14.17)

PDF only: Make a full copy (duplicate) of a page.

  • pno (int) – the page to be duplicated. Must be in range 0 <= pno < page_count.
  • to (int) – the page number in front of which to copy. The default inserts after the last page.


  • In contrast to copy_page(), this method creates a new page object (with a new xref), which can be changed independently from the original.
  • Any Popup and “IRT” (“in response to”) annotations are not copied to avoid potentially incorrect situations.
move_page(pno, to=-1)

PDF only: Move (copy and then delete original) a page within the document.

  • pno (int) – the page to be moved. Must be in range 0 <= pno < page_count.
  • to (int) – the page number in front of which to insert the moved page. The default moves after the last page.

(New in v1.17.4)

PDF only: Get or set the /NeedAppearances property of Form PDFs. Quote: “(Optional) A flag specifying whether to construct appearance streams and appearance dictionaries for all widget annotations in the document … Default value: false.” This may help controlling the behavior of some readers / viewers.

Parameters:value (bool) – set the property to this value. If omitted or None, inquire the current value.
Return type:bool
  • None: not a Form PDF, or property not defined.
  • True / False: the value of the property (either just set or existing for inquiries). Has no effect if no Form PDF.

PDF only: Return whether the document contains signature fields. This is an optional PDF property: if not present (return value -1), no conclusions can be drawn – the PDF creator may just not have bothered using it.

Return type:int
  • -1: not a Form PDF / no signature fields recorded / no SigFlags found.
  • 1: at least one signature field exists.
  • 3: contains signatures that may be invalidated if the file is saved (written) in a way that alters its previous contents, as opposed to an incremental update.
embfile_add(name, buffer, filename=None, ufilename=None, desc=None)

PDF only: Embed a new file. All string parameters except the name may be unicode (in previous versions, only ASCII worked correctly). File contents will be compressed (where beneficial).

Changed in version 1.14.16
The sequence of positional parameters “name” and “buffer” has been changed to comply with the layout of other functions.
  • name (str) – entry identifier, must not already exist.
  • buffer (bytes,bytearray,BytesIO) –

    file contents.

    (Changed in version 1.14.13) io.BytesIO is now also supported.

  • filename (str) – optional filename. Documentation only, will be set to name if None.
  • ufilename (str) – optional unicode filename. Documentation only, will be set to filename if None.
  • desc (str) – optional description. Documentation only, will be set to name if None.
Return type:



(Changed in v1.18.13) The method now returns the xref of the inserted file. In addition, the file object now will be automatically given the PDF keys /CreationDate and /ModDate based on the current date-time.


PDF only: Return the number of embedded files.

Changed in version 1.14.16
This is now a method. In previous versions, this was a property.

PDF only: Retrieve the content of embedded file by its entry number or name. If the document is not a PDF, or entry cannot be found, an exception is raised.

Parameters:item (int,str) – index or name of entry. An integer must be in range(embfile_count()).
Return type:bytes

PDF only: Remove an entry from /EmbeddedFiles. As always, physical deletion of the embedded file content (and file space regain) will occur only when the document is saved to a new file with a suitable garbage option.

Changed in version 1.14.16
Items can now be deleted by index, too.
Parameters:item (int/str) – index or name of entry.


When specifying an entry name, this function will only delete the first item with that name. Be aware that PDFs not created with PyMuPDF may contain duplicate names. So you may want to take appropriate precautions.


(Changed in v1.18.13)

PDF only: Retrieve information of an embedded file given by its number or by its name.

Parameters:item (int/str) – index or name of entry. An integer must be in range(embfile_count()).
Return type:dict
Returns:a dictionary with the following keys:
  • name – (str) name under which this entry is stored
  • filename – (str) filename
  • ufilename – (unicode) filename
  • desc – (str) description
  • size – (int) original file size
  • length – (int) compressed file length
  • creationDate(New in v1.18.13) (str) date-time of item creation in PDF format
  • modDate(New in v1.18.13) (str) date-time of last change in PDF format
  • collection(New in v1.18.13) (int) xref of the associated PDF portfolio item if any, else zero.
  • checksum(New in v1.18.13) (str) a hashcode of the stored file content as a hexadecimal string. Should be MD5 according to PDF specifications, but be prepared to see other hashing algorithms.

(New in version 1.14.16)

PDF only: Return a list of embedded file names. The sequence of the names equals the physical sequence in the document.

Return type:list
embfile_upd(item, buffer=None, filename=None, ufilename=None, desc=None)

PDF only: Change an embedded file given its entry number or name. All parameters are optional. Letting them default leads to a no-operation.

  • item (int/str) – index or name of entry. An integer must be in range(embfile_count()).
  • buffer (bytes,bytearray,BytesIO) –

    the new file content.

    (Changed in version 1.14.13) io.BytesIO is now also supported.

  • filename (str) – the new filename.
  • ufilename (str) – the new unicode filename.
  • desc (str) – the new description.

(Changed in v1.18.13) The method now returns the xref of the file object.

Return type:int
Returns:xref of the file object. Automatically, its /ModDate PDF key will be updated with the current date-time.

Release objects and space allocations associated with the document. If created from a file, also closes filename (releasing control to the OS). Explicitely closing a document is equivalent to deleting it, del doc, or assigning it to something else like doc = None.

xref_object(xref, compressed=False, ascii=False)

(New in version 1.16.8, changed in v1.18.10)

PDF only: Return the definition source of a PDF object.

  • xref (int) – the object’s :data`xref`. Changed in v1.18.10: A value of -1 returns the PDF trailer source.
  • compressed (bool) – whether to generate a compact output with no line breaks or spaces.
  • ascii (bool) – whether to ASCII-encode binary data.
Return type:



The object definition source.


(New in version 1.16.8)

PDF only: Return the xref number of the PDF catalog (or root) object. Use that number with Document.xref_object() to see its source.


(New in version 1.16.8)

PDF only: Return the trailer source of the PDF, which is usually located at the PDF file’s end. This is Document.xref_object() with an xref argument of -1.


PDF Only: Extract data and meta information of an image stored in the document. The output can directly be used to be stored as an image file, as input for PIL, Pixmap creation, etc. This method avoids using pixmaps wherever possible to present the image in its original format (e.g. as JPEG).

Parameters:xref (int) – xref of an image object. If this is not in range(1, doc.xref_length()), or the object is no image or other errors occur, None is returned and no exception is raised.
Return type:dict
Returns:a dictionary with the following keys
  • ext (str) image type (e.g. ‘jpeg’), usable as image file extension
  • smask (int) xref number of a stencil (/SMask) image or zero
  • width (int) image width
  • height (int) image height
  • colorspace (int) the image’s colorspace.n number.
  • cs-name (str) the image’s
  • xres (int) resolution in x direction. Please also see resolution.
  • yres (int) resolution in y direction. Please also see resolution.
  • image (bytes) image data, usable as image file content
>>> d = doc.extract_image(1373)
>>> d
{'ext': 'png', 'smask': 2934, 'width': 5, 'height': 629, 'colorspace': 3, 'xres': 96,
'yres': 96, 'cs-name': 'DeviceRGB',
'image': b'\x89PNG\r\n\x1a\n\x00\x00\x00\rIHDR\x00\x00\x00\x05\ ...'}
>>> imgout = open("image." + d["ext"], "wb")
>>> imgout.write(d["image"])
>>> imgout.close()


There is a functional overlap with pix = fitz.Pixmap(doc, xref), followed by a pix.tobytes(). Main differences are that extract_image, (1) does not always deliver PNG image formats, (2) is very much faster with non-PNG images, (3) usually results in much less disk storage for extracted images, (4) returns None in error cases (generates no exception). Look at the following example images within the same PDF.

  • xref 1268 is a PNG – Comparable execution time and identical output:

    In [23]: %timeit pix = fitz.Pixmap(doc, 1268);pix.tobytes()
    10.8 ms ± 52.4 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
    In [24]: len(pix.tobytes())
    Out[24]: 21462
    In [25]: %timeit img = doc.extract_image(1268)
    10.8 ms ± 86 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
    In [26]: len(img["image"])
    Out[26]: 21462
  • xref 1186 is a JPEG – Document.extract_image() is many times faster and produces a much smaller output (2.48 MB vs. 0.35 MB):

    In [27]: %timeit pix = fitz.Pixmap(doc, 1186);pix.tobytes()
    341 ms ± 2.86 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
    In [28]: len(pix.tobytes())
    Out[28]: 2599433
    In [29]: %timeit img = doc.extract_image(1186)
    15.7 µs ± 116 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each)
    In [30]: len(img["image"])
    Out[30]: 371177
extract_font(xref, info_only=False, named=None)
  • Changed in v1.19.4: return a dictionary if named == True.

PDF Only: Return an embedded font file’s data and appropriate file extension. This can be used to store the font as an external file. The method does not throw exceptions (other than via checking for PDF and valid xref).

arg int xref:

PDF object number of the font to extract.

arg bool info_only:

only return font information, not the buffer. To be used for information-only purposes, avoids allocation of large buffer areas.

arg bool named:

If true, a dictionary with the following keys is returned: ‘name’ (font base name), ‘ext’ (font file extension), ‘type’ (font type), ‘content’ (font file content).




a tuple (basename, ext, type, content), where ext is a 3-byte suggested file extension (str), basename is the font’s name (str), type is the font’s type (e.g. “Type1”) and content is a bytes object containing the font file’s content (or b””). For possible extension values and their meaning see Font File Extensions. Return details on error:

  • ("", "", "", b"") – invalid xref or xref is not a (valid) font object.
  • (basename, "n/a", "Type1", b"")basename is not embedded and thus cannot be extracted. This is the case for e.g. the PDF Base 14 Fonts and Type 3 fonts.


>>> # store font as an external file
>>> name, ext, _, content = doc.extract_font(4711)
>>> # assuming content is not None:
>>> ofile = open(name + "." + ext, "wb")
>>> ofile.write(content)
>>> ofile.close()


The basename is returned unchanged from the PDF. So it may contain characters (such as blanks) which may disqualify it as a filename for your operating system. Take appropriate action.


  • The returned basename in general is not the original file name, but it probably has some similarity.
  • If parameter named == True, a dictionary with the following keys is returned: {'name': 'T1', 'ext': 'n/a', 'type': 'Type3', 'content': b''}.

(New in version 1.16.8)

PDF only: Return the xref of the document’s XML metadata.


(New in version 1.16.8)

PDF only: Return the decompressed contents of the xref stream object.

Parameters:xref (int) – xref number.
Return type:bytes
Returns:the (decompressed) stream of the object.

(New in version 1.16.8)

PDF only: Return the unmodified (esp. not decompressed) contents of the xref stream object. Otherwise equal to Document.xref_stream().

Return type:bytes
Returns:the (original, unmodified) stream of the object.
update_object(xref, obj_str, page=None)

(New in version 1.16.8)

PDF only: Replace object definition of xref with the provided string. The xref may also be new, in which case this instruction completes the object definition. If a page object is also given, its links and annotations will be reloaded afterwards.

  • xref (int) – xref number.
  • obj_str (str) – a string containing a valid PDF object definition.
  • page (Page) – a page object. If provided, indicates, that annotations of this page should be refreshed (reloaded) to reflect changes incurred with links and / or annotations.
Return type:



zero if successful, otherwise an exception will be raised.

update_stream(xref, data, new=False, compress=True)
  • New in v.1.16.8
  • Changed in v1.19.2: added parameter “compress”

Replace the stream of an object identified by xref. If the object has no stream, an exception is raised unless new=True is used. The function automatically performs a compress operation (“deflate”) where beneficial.

  • xref (int) – xref number.
  • stream (bytes|bytearray|BytesIO) –

    the new content of the stream.

    (Changed in version 1.14.13:) io.BytesIO objects are now also supported.

  • new (bool) – whether to force accepting the stream, and thus turning it into a stream object.
  • compress (bool) –

    whether to compress the inserted stream. If True (default), the stream will be inserted using /FlateDecode compression, otherwise the stream will inserted as is.


    The object of xref must be a PDF dictionary for this to work, and especially must not be empty – as is the case if you just created the xref via Document.get_new_xref(). To avoid this, at a minimum execute doc.update_object(xref, "<<>>") before inserting the stream.

This method is primarily intended to manipulate streams containing PDF operator syntax (see pp. 643 of the Adobe PDF References) as it is the case for e.g. page content streams.

If you update a contents stream, you should use save parameter clean=True. This ensures consistency between PDF operator source and the object structure.

Example: Let us assume that you no longer want a certain image appear on a page. This can be achieved by deleting the respective reference in its contents source(s) – and indeed: the image will be gone after reloading the page. But the page’s resources object would still show the image as being referenced by the page. This save option will clean up any such mismatches.


(New in v1.18.7)

PDF only: Check whether there are links, resp. annotations anywhere in the document.

Returns:True / False. As opposed to fields, which are also stored in a central place of a PDF document, the existence of links / annotations can only be detected by parsing each page. These methods are tuned to do this efficiently and will immediately return, if the answer is True for a page. For PDFs with many thousand pages however, an answer may take some time [6] if no link, resp. no annotation is found.

(New in v1.18.7, changed in v1.18.9)

PDF only: Investigate eligible fonts for their use by text in the document. If a font is supported and a size reduction is possible, that font is replaced by a version with a character subset.

Use this method immediately before saving the document. The following features and restrictions apply for the time being:

  • Package fontTools must be installed. It is required for creating the font subsets. If not installed, the method raises an ImportError exception.
  • Supported font types only include embedded OTF, TTF and WOFF that are not already subsets.
  • The script directory must be available for writing temporary files during the subsetting process.
  • Changed in v1.18.9: A subset font directly replaces its original – text remains untouched and is not rewritten. It thus should retain all its properties, like spacing, hiddenness, control by Optional Content, etc.

The greatest benefit can be achieved when creating new PDFs using large fonts like is typical for Asian scripts. In these cases, the set of actually used unicodes mostly is small compared to the number of glyphs in the font. Using this feature can easily reduce the embedded font binary by two orders of magnitude – from several megabytes to a low two-digit kilobyte amount.

  • New in v1.19.0

PDF only: Enable journalling. Use this before you start logging operations.

  • New in v1.19.0

PDF only: Start journalling an “operation” identified by a string “name”. Updates will fail for a journal-enabled PDF, if no operation has been started.

  • New in v1.19.0

PDF only: Stop the current operation. The updates between start and stop of an operation belong to the same unit of work and will be undone / redone together.

  • New in v1.19.0

PDF only: Return the numbers of the current operation and the total operation count.

Returns:a tuple (step, steps) containing the current operation number and the total number of operations in the journal. If step is 0, we are at the top of the journal. If step equals steps, we are at the bottom. Updating the PDF with anything other than undo or redo will automatically remove all journal entries after the current one and the new update will become the new last entry in the journal. The updates corresponding to the removed journal entries will be permanently lost.
  • New in v1.19.0

PDF only: Return the name of operation number step.

  • New in v1.19.0

PDF only: Show whether forward (“redo”) and / or backward (“undo”) executions are possible from the current journal postion.

Returns:a dictionary {"undo": bool, "redo": bool}. The respective method is available if its value is True.
  • New in v1.19.0

PDF only: Revert (undo) the current step in the journal. This moves towards the journal’s top.

  • New in v1.19.0

PDF only: Re-apply (redo) the current step in the journal. This moves towards the journal’s bottom.

  • New in v1.19.0

PDF only: Save the journal to a file.

Parameters:filename (str,fp) – either a filename as string or a file object opened as “wb” (or an io.BytesIO() object).
  • New in v1.19.0

PDF only: Load journal from a file. Enables journalling for the document. If journalling is already enabled, an exception is raised.

Parameters:filename (str,fp) – the filename (str) of the journal or a file object opened as “rb” (or an io.BytesIO() object).
  • New in v1.19.0

PDF only: Saves a “snapshot” of the document. This is a PDF document with a special, incremental-save format compatible with journalling – therefore no save options are available. Saving a snapshot is not possible for new documents.

This is a normal PDF document with no usage restrictions whatsoever. If it is not being changed in any way, it can be used together with its journal to undo / redo operations or continue updating.


Contains the first Outline entry of the document (or None). Can be used as a starting point to walk through all outline items. Accessing this property for encrypted, not authenticated documents will raise an AttributeError.


False if document is still open. If closed, most other attributes and methods will have been deleted / disabled. In addition, Page objects referring to this document (i.e. created with Document.load_page()) and their dependent objects will no longer be usable. For reference purposes, still exists and will contain the filename of the original document (if applicable).


True if this is a PDF document and contains unsaved changes, else False.


True if this is a PDF document, else False.


False if this is not a PDF or has no form fields, otherwise the number of root form fields (fields with no ancestors).

(Changed in version 1.16.4) Returns the total number of (root) form fields.


True if document has a variable page layout (like e-books or HTML). In this case you can set the desired page dimensions during document creation (open) or via method layout().


(New in v1.18.2)

True if PDF has been repaired during open (because of major structure issues). Always False for non-PDF documents. If true, more details have been stored in TOOLS.mupdf_warnings(), and Document.can_save_incrementally() will return False.


Indicates whether the document is password-protected against access. This indicator remains unchanged – even after the document has been authenticated. Precludes incremental saves if true.


This indicator initially equals Document.needs_pass. After successful authentication, it is set to False to reflect the situation.


Contains the permissions to access the document. This is an integer containing bool values in respective bit positions. For example, if doc.permissions & fitz.PDF_PERM_MODIFY > 0, you may change the document. See Document Permissions for details.

Changed in version 1.16.0 This is now an integer comprised of bit indicators. Was a dictionary previously.


Contains the document’s meta data as a Python dictionary or None (if is_encrypted=True and needPass=True). Keys are format, encryption, title, author, subject, keywords, creator, producer, creationDate, modDate, trapped. All item values are strings or None.

Except format and encryption, for PDF documents, the key names correspond in an obvious way to the PDF keys /Creator, /Producer, /CreationDate, /ModDate, /Title, /Author, /Subject, /Trapped and /Keywords respectively.

  • format contains the document format (e.g. ‘PDF-1.6’, ‘XPS’, ‘EPUB’).

  • encryption either contains None (no encryption), or a string naming an encryption method (e.g. ‘Standard V4 R4 128-bit RC4’). Note that an encryption method may be specified even if needs_pass=False. In such cases not all permissions will probably have been granted. Check Document.permissions for details.

  • If the date fields contain valid data (which need not be the case at all!), they are strings in the PDF-specific timestamp format “D:<TS><TZ>”, where

    • <TS> is the 12 character ISO timestamp YYYYMMDDhhmmss (YYYY - year, MM - month, DD - day, hh - hour, mm - minute, ss - second), and
    • <TZ> is a time zone value (time intervall relative to GMT) containing a sign (‘+’ or ‘-‘), the hour (hh), and the minute (‘mm’, note the apostrophies!).
  • A Paraguayan value might hence look like D:20150415131602-04’00’, which corresponds to the timestamp April 15, 2015, at 1:16:02 pm local time Asuncion.


Contains the filename or filetype value with which Document was created.


Contains the number of pages of the document. May return 0 for documents with no pages. Function len(doc) will also deliver this result.


(New in version 1.17.0) Contains the number of chapters in the document. Always at least 1. Relevant only for document types with chapter support (EPUB currently). Other documents will return 1.


(New in version 1.17.0) Contains (chapter, pno) of the document’s last page. Relevant only for document types with chapter support (EPUB currently). Other documents will return (0, len(doc) - 1) and (0, -1) if it has no pages.


A list of form field font names defined in the /AcroForm object. None if not a PDF.



For methods that change the structure of a PDF (insert_pdf(), select(), copy_page(), delete_page() and others), be aware that objects or properties in your program may have been invalidated or orphaned. Examples are Page objects and their children (links, annotations, widgets), variables holding old page counts, tables of content and the like. Remember to keep such variables up to date or delete orphaned objects. Also refer to Ensuring Consistency of Important Objects in PyMuPDF.

set_metadata() Example

Clear metadata information. If you do this out of privacy / data protection concerns, make sure you save the document as a new file with garbage > 0. Only then the old /Info object will also be physically removed from the file. In this case, you may also want to clear any XML metadata inserted by several PDF editors:

>>> import fitz
>>> doc.metadata             # look at what we currently have
{'producer': 'rst2pdf, reportlab', 'format': 'PDF 1.4', 'encryption': None, 'author':
'Jorj X. McKie', 'modDate': "D:20160611145816-04'00'", 'keywords': 'PDF, XPS, EPUB, CBZ',
'title': 'The PyMuPDF Documentation', 'creationDate': "D:20160611145816-04'00'",
'creator': 'sphinx', 'subject': 'PyMuPDF 1.9.1'}
>>> doc.set_metadata({})      # clear all fields
>>> doc.metadata             # look again to show what happened
{'producer': 'none', 'format': 'PDF 1.4', 'encryption': None, 'author': 'none',
'modDate': 'none', 'keywords': 'none', 'title': 'none', 'creationDate': 'none',
'creator': 'none', 'subject': 'none'}
>>> doc._delXmlMetadata()    # clear any XML metadata
>>>"anonymous.pdf", garbage = 4)       # save anonymized doc

set_toc() Demonstration

This shows how to modify or add a table of contents. Also have a look at and in the examples directory.

>>> import fitz
>>> doc ="test.pdf")
>>> toc = doc.get_toc()
>>> for t in toc: print(t)                           # show what we have
[1, 'The PyMuPDF Documentation', 1]
[2, 'Introduction', 1]
[3, 'Note on the Name fitz', 1]
[3, 'License', 1]
>>> toc[1][1] += " modified by set_toc"               # modify something
>>> doc.set_toc(toc)                                  # replace outline tree
3                                                    # number of bookmarks inserted
>>> for t in doc.get_toc(): print(t)                  # demonstrate it worked
[1, 'The PyMuPDF Documentation', 1]
[2, 'Introduction modified by set_toc', 1]            # <<< this has changed
[3, 'Note on the Name fitz', 1]
[3, 'License', 1]

insert_pdf() Examples

(1) Concatenate two documents including their TOCs:

>>> doc1 ="file1.pdf")          # must be a PDF
>>> doc2 ="file2.pdf")          # must be a PDF
>>> pages1 = len(doc1)                     # save doc1's page count
>>> toc1 = doc1.get_toc(False)     # save TOC 1
>>> toc2 = doc2.get_toc(False)     # save TOC 2
>>> doc1.insert_pdf(doc2)                   # doc2 at end of doc1
>>> for t in toc2:                         # increase toc2 page numbers
        t[2] += pages1                     # by old len(doc1)
>>> doc1.set_toc(toc1 + toc2)               # now result has total TOC

Obviously, similar ways can be found in more general situations. Just make sure that hierarchy levels in a row do not increase by more than one. Inserting dummy bookmarks before and after toc2 segments would heal such cases. A ready-to-use GUI (wxPython) solution can be found in script of the examples directory.

(2) More examples:

>>> # insert 5 pages of doc2, where its page 21 becomes page 15 in doc1
>>> doc1.insert_pdf(doc2, from_page=21, to_page=25, start_at=15)
>>> # same example, but pages are rotated and copied in reverse order
>>> doc1.insert_pdf(doc2, from_page=25, to_page=21, start_at=15, rotate=90)
>>> # put copied pages in front of doc1
>>> doc1.insert_pdf(doc2, from_page=21, to_page=25, start_at=0)

Other Examples

Extract all page-referenced images of a PDF into separate PNG files:

for i in range(len(doc)):
    imglist = doc.get_page_images(i)
    for img in imglist:
        xref = img[0]                  # xref number
        pix = fitz.Pixmap(doc, xref)   # make pixmap from image
        if pix.n - pix.alpha < 4:      # can be saved as PNG
  "p%s-%s.png" % (i, xref))
        else:                          # CMYK: must convert first
            pix0 = fitz.Pixmap(fitz.csRGB, pix)
  "p%s-%s.png" % (i, xref))
            pix0 = None                # free Pixmap resources
        pix = None                     # free Pixmap resources

Rotate all pages of a PDF:

>>> for page in doc: page.set_rotation(90)


[1]Content streams describe what (e.g. text or images) appears where and how on a page. PDF uses a specialized mini language similar to PostScript to do this (pp. 643 in Adobe PDF References), which gets interpreted when a page is loaded.
[2]However, you can use Document.get_toc() and Page.get_links() (which are available for all document types) and copy this information over to the output PDF. See demo
[3]For applicable (EPUB) document types, loading a page via its absolute number may result in layouting a large part of the document, before the page can be accessed. To avoid this performance impact, prefer chapter-based access. Use convenience methods and attributes Document.next_location(), Document.prev_location() and Document.last_location for maintaining a high level of coding efficiency.
[4](1, 2) These parameters cause separate handling of stream categories: use it together with expand to restrict decompression to streams other than images / fontfiles.
[5]Examples for “Form XObjects” are created by Page.show_pdf_page().
[6]For a False the complete document must be scanned. Both methods do not load pages, but only scan object definitions. This makes them at least 10 times faster than application-level loops (where total response time roughly equals the time for loading all pages). For the Adobe PDF References (756 pages) and the Pandas documentation (over 3070 pages) – both have no annotations – the method needs about 11 ms for the answer False. So response times will probably become significant only well beyond this order of magnitude.
[7]This only works under certain conditions. For example, if there is normal text covered by some image on top of it, then this is undetectable and the respective text is not removed. Similar is true for white text on white background, and so on.