Pure Python library for saving and loading PNG images

png, python
pip install pypng==0.20220715.0




PYPNG HAS MOVED TO GITLAB: https://gitlab.com/drj11/pypng


PNG module for Python. PyPNG is written entirely in Python.


import png
png.from_array([[255, 0, 0, 255],
                [0, 255, 255, 0]], 'L').save("small_smiley.png")

After that, try import png then help(png). Also, lickable HTML documentation appears in the html/ directory. If HTML is no good then you could try the ReST sources in the man/ directory.


PyPNG is pure Python and has no dependencies. It requires Python 3.5 or any compatible higher version.

To install PyPNG package via pip use:

python -m pip install git+https://gitlab.com/drj11/pypng@pypng-0.0.21

You can also use setuptools to install from source; PyPNG uses setup.cfg and pyproject.toml to record its configuration.

To install from (version controlled) sources using a suitable version of pip:

cd into the directory and execute the command:

python -m pip install .

The png module will be installed; import png will allow you to use it from your Python programs.

PyPNG is so simple, that you don't need installation tools. You can copy code/png.py wherever you like. It's intended that you can copy png.py into your application and distribute it. The following curl command should copy the latest version into your current directory:

curl -LO https://gitlab.com/drj11/pypng/-/raw/main/code/png.py


(For issues see https://gitlab.com/drj11/pypng/-/issues/ )

Release (the next)

Development moved to gitlab: https://gitlab.com/drj11/pypng

If you pass an empty file to PyPNG it now raises the builtin Python exception EOFError. This should make it easier to diagnose empty file problems separately from genuine format errors (which use png.FormatError). This is a slightly breaking change to the API.

New prirowpng tool to join PNG images in a row left-to-right (old internal pipcat tool).

New pricolpng tool to join PNG images in a column top-to-bottom.

Support for plain PGM files (magic number P2) added to pripamtopng.

New priplan9topng tool to convert from Plan 9 image format to PNG. In reality this has been lurking in the codebase for years, but has recently been converted to Python 3. The author has only a limited collection of Plan 9 images, which limits the testing that can be done. The author welcomes bug reports for Plan 9 images.

The priplan9topng tool has an even more experimental option --font which converts Plan 9 subfont files to a sequence of PNG files.

Release 0.0.21

Support for Python 2 is dropped. Python 3.5 and onwards are supported. Some of the ancillary tools are modified to work on Python 3.

Installs via wheel files.

prichunkpng command line tool now has some new options to add chunks:

  • --iccprofile to add a iCCP chunk (ICC Profile);
  • --physical to add a pHYs chunk, specifying the intended pixel size;
  • --sigbit to add a sBIT chunk, specifying the encoded significant bits;
  • --transparent to add a tRNS chunk, specifying the transparent colour.

priditherpng command line tool standardised and converted to Python 3.

pripngtopam tool now has a --plain option to output plain PGM and PPM formats. The topam part of the name is a bit of a misnomer: when possible (L and RGB PNG files) the tool will output either a PGM (grey) or a PPM (RGB) file. Essentially all tools that can process a PAM file can also process a PGM or a PPM file. PAM files cannot be plain so using the option will raise an error in the case where a true PAM file is written.

Better error messages when you write the wrong number of rows.

(Slightly experimentally) running the png module as a command line tool, with python -m png, will report the version and file location of the png module.

Release 0.0.20

Support for earlier versions of Python is dropped. Python 3.4 and onwards are supported. Python 2.7 also works, but this is the last release to support any version of Python 2.

Cython code is removed, which simplifies the implementation.

Removed the (optional) dependency setuptools.

Changed the default for png.Writer to be greyscale.

Removed 3D support from .from_array().

Release 0.0.19

Support for earlier versions of Python is dropped in order to simplify the code. From the Python 3.x series all versions from 3.2 onward are supported. From the 2.x series only Python 2.6 and 2.7 are supported.

Code cleaned. Tests renamed and more organised. Generally Flake 8 compliant. Fewer special cases for ancient versions of Python.

The row length is checked when writing PNG files. Previously it was possible to write badly formed PNG files by passing in rows of the wrong length.

pHYS chunk is now processed.

The Writer() interface now supports source pixels that have a different bitdepth for each channel. To exploit this, pass in a tuple for the bitdepth argument.

Ancillary tools regularised and simplified.

pripamtopng command line tool converts NetPBM PNM/PAM files to PNG. Previously png.py did this. Note that only one input file is converted, if you have intensity and opacity in separate files, you have two options: either use pamstack to convert them into a single PAM file and convert that, or convert each file to PNG, then use priweavepng to weave them together. Both will work.

pripngtopam command line tool converts PNG to NetPBM PNM/PAM file. Previously png.py did this.

python -m pngsuite is now a command line tool to write various images from the PNG suite of test images. Previously this was possible using gen.

priweavepng command line tool performs channel extraction across multiple images. This is a more general version of pipstack (which has been removed), and is inspired by pamstack from NetPBM.

The --interlace option available on many previous tools is now only available on priweavepng, making it the preferred tool for generating interlaced PNGs.

prichunkpng command line tool adds and deletes chunks and "intelligently" knows about transparent, gamma, background chunks. It is the preferred tool for adding those chunks, which was previously possible using various options of other tools.

gen has been renamed to priforgepng.

priforgepng uses centre pixel sampling, which means that a 256 pixel wide 8-bit gradient takes on all possible 256 values. It also improves output for very small images.

priforgepng uses Fraction for internal maths meaning that the stripe patterns are accurate and do not have loose pixels.

priforgepng only outputs greyscale PNG files (but see next item for a feature to generate colour PNGs).

priforgepng can output multiple PNGs onto the same stream. This aligns well with a feature of priweavepng which accepts multiple PNGs from stdin. LA, RGB, and RGBA test images can be generated by piping priforgepng into priweavepng: priforgepng RTL RTR RBR | priweavepng - - - will generate an RGB PNG.

Release 0.0.18

Thanks to github.com/sean-duffy, .from_array() can now take a 3D array.

Converting to PNMs was broken in Python 3; this is now fixed. Issue 26: https://gitlab.com/drj11/pypng/-/issues/26

Release 0.0.17

Various fixes when running on Python 3 and Windows. Merging pull requests from github.com/ironfroggy and github.com/techtonik, and merging by hand a commit from github.com/Scondo.

Release 0.0.16

Compatible with nose: nosetests png.py now works.

Allow any "file-like" object as an input.

Handle newlines in texttopng.

Release 0.0.15

Fixed various URLs to point at github.com instead of googlecode.

Release 0.0.14

When using png.py as a command line tool, it can now produce non-square test images.

PyPNG now installs "out of the box" on Python 3 on a plain install (previously distribute or pip was required).

PyPNG welcomes the following community contributions:

Joaquín Cuenca Abela speeds up PNG reading when Cython is available.

Josh Bleecher Snyder adds a lenient mode which has relaxed checksum checking.

nathan@dunfield.info fixed a problem writing files when using the command line tool on Windows (Issue 62).

The following issues have been fixed:

On github:

Issue 6: Palette processing is annoying

On googlecode:

Issue 62: Problem writing PNG files on Windows

Development has moved from googlecode to github. All issues below here, and the one immediately above, are from the googlecode issue tracker. All newer issue should be on github.

Release 0.0.13

PyPNG now installs "out of the box" on Python 3. Thanks to simon.sapin@kozea.fr and nathan@dunfield.info for the patch.

The following issues have been fixed:

Issue 63: setup.py does not use 2to3 Issue 64: Typo in documentation

Release 0.0.12

PyPNG now works on Python 3 if you use the 2to3 tool. Fix for converting grey images to RGBA.

The following issues have been fixed:

Issue 60: Greyscale images not properly being converted to RGBA Issue 61: Doesn't work on Python 3

Release 0.0.11

Added the "How Fast is PyPNG" section to the documentation. Changed it so that more PNG formats return their rows as Python array.array instances.

Release 0.0.10

Fix for read_flat method (broken for ages).

The following issues have been fixed:

Issue 56: read_flat broken

Release 0.0.9

Tentative fix for a deprecation warning on 64-bit Python 2.5 systems. Conversion tool for Plan 9 images.

Issue 54 (below) is tentative. The PyPNG developers have been unable to reproduce the error (as it seems to be on 64-bit Python 2.5 systems); any user reports would be most welcome.

The following issues have been fixed:

Issue 54: Deprecation warnings when using pypng. Issue 55: Cannot convert Plan 9 images.

Release 0.0.8

Mostly more robust to dodgy input PNGs, as a result of testing with brokensuite. One fixed bug was a critical: an infinite loop for a least one input (Issue 52 below).

The following issues have been fixed:

Issue 47: Leading blanks when using write_packed. Issue 48: pipdither fails when input has no gamma chunk. Issue 49: pipdither fail with 1-bit input. Issue 50: pipchunk adds second gamma chunk. Issue 51: piprgb and pipasgrey fail for color mapped images. Issue 52: Some inputs cause infinite loop.

Release 0.0.7

Better documentation (in html/ex.html mostly) for NumPy integration.

The following issues have been fixed:

Issue 46: Unclear how to get PNG pixel data into and out of NumPy.

Release 0.0.6

NumPy integer types now work.

The following issues have been fixed:

Issue 44: Cannot use numpy.uint16 for pixel values.

Release 0.0.5

sBIT chunks are now handled, meaning that PyPNG can handle any (single) bit depth from 1 to 16 from end to end.

The following issues have been fixed:

Issue 28: Does not add sBIT chunk. Issue 36: Ignores sBIT chunk when present.

Release 0.0.4

PyPNG now works on Python 2.2 (significant for Symbian users as PyS60 is based on Python 2.2). Not all features are supported on Python 2.2.

The following issues have been fixed:

Issue 16: Source and doc uses 'K' where it should use 'L'. Issue 32: Does not accept packed data. Issue 33: Cannot create greyscale PNG with transparency. Issue 35: Does not work on Python 2.2.

Release 0.0.3

Handling PAM files allows end to end handling of alpha channels in workflows that involve both Netpbm formats and PNG. PyPNG now works in Python 2.3.

The following issues have been fixed:

Issue 14: Does not read PAM files. Issue 15: Does not write PAM files. Issue 25: Incorrect handling of tRNS chunk. Issue 26: asRGBA8 method crashes out for color type 2 images. Issue 27: Fails on Python 2.3.

Release 0.0.2

Lickable HTML documentation is now provided (see the html/ directory), generated by Sphinx.

The following issues have been fixed:

Issue 8: Documentation is not lickable. Issue 9: Advantage over PIL is not clear. Issue 19: Bogus message for PNM inputs with unsupported maxval Issue 20: Cannot write large PNG files

Release 0.0.1

Stuff happened.


  • .../ - top-level crud (like this README, and setup.py).
  • .../asset - assets (needed for testing)
  • .../code/ - the Python code.
  • .../man/ - manuals (in source/plain-text).
  • .../proc/ - documented procedures (release procedure).