13.0.0

Security

CVE identifiers will be added to these entries as they are assigned.

Critical

High

Medium

Low

Backwards incompatible changes

Python 3.10

Pillow has dropped support for Python 3.10, which reached end-of-life in October 2026.

Saving I mode images as PNG

In order to fit the 32 bits of I mode images into PNG, when PNG images can only contain at most 16 bits for a channel, Pillow has been clipping the values. Rather than quietly changing the data, this is now removed. Instead, the image can be converted to another mode before saving:

from PIL import Image
im = Image.new("I", (1, 1))
im.convert("I;16").save("out.png")

Image.fromarray mode parameter

Using the mode parameter in fromarray() was deprecated in Pillow 11.3.0. In Pillow 12.0.0, this was partially reverted, and now the only functionality removed is when the mode changes data types. Since pixel values do not contain information about palettes or color spaces, the parameter can still be used to place grayscale L mode data within a P mode image, or read RGB data as YCbCr for example. If omitted, the mode will be automatically determined from the object’s shape and type.

Image.Image.get_child_images()

Image.Image.get_child_images() has been moved to ImageFile.ImageFile.get_child_images(). The method uses an image’s file pointer, and so child images could only be retrieved from an PIL.ImageFile.ImageFile instance.

Image._show

Image._show has been removed. Use show() instead.

ImageCms.ImageCmsProfile.product_name and .product_info

ImageCms.ImageCmsProfile.product_name and the corresponding .product_info attributes have been removed. They were set to None since Pillow 2.3.0.

ExifTags.IFD.Makernote

ExifTags.IFD.Makernote has been removed. Instead, use ExifTags.IFD.MakerNote.

Deprecations

IM image format

The IM image format has been deprecated and removal is planned in Pillow 15.0.0 (2028-10-15). If you are using this format and would like to continue doing so, open an issue about what other software uses it.

JpegImageFile.load_djpeg

JpegImageFile.load_djpeg has been deprecated, and will be removed in Pillow 14 (2027-10-15).

Use the built-in JPEG decoder instead, or call djpeg directly and decode the resulting image with Pillow.

ImageQt align8to32()

ImageQt.align8to32() has been deprecated. This was an undocumented helper function intended for internal use, so there is no replacement.

API changes

ImageColor.getcolor() with premultiplied alpha modes

When mode is RGBa or La, getcolor() now returns color values premultiplied by the alpha channel, matching how pixels are stored in those modes. Previously, the unmultiplied values were returned.

Duplicate TIFF tags

If a TIFF IFD contains the same tag more than once, the first value is now kept. Previously, the last value overwrote the earlier ones.

Saving small ICO images

When saving an image smaller than 16x16 pixels as an ICO without a sizes argument, the image’s own size is now used. Previously, none of the default sizes matched, and an ICO containing no images was written. If no size in sizes fits within the image, a ValueError is now raised.

API additions

New resizing filters

Two new filters are available for resize() and thumbnail(): Image.Resampling.MKS2013 and Image.Resampling.MKS2021. These are versions of the Magic Kernel Sharp filter.

Added strip_namespaces argument to Image.getxmp()

getxmp() now accepts an optional keyword argument of strip_namespaces. It is set to True by default, stripping each tag’s XML namespace URI prefix as before. If set to False, each tag’s full {namespace-uri}local-name form is kept instead, avoiding collisions between tags that share a local name across different namespaces.

Added escape argument to PSDraw.text()

text() now accepts a keyword-only boolean argument, escape. Set it to True to escape backslashes, for example in Windows paths. The default, False, preserves the existing handling of backslashes and PostScript escape sequences.

ImageOps invert, posterize and solarize of P images

invert(), posterize() and solarize() now support P mode images. The operation is applied to the palette colors, leaving any alpha values in the palette unchanged. Previously, these functions raised NotImplementedError for P mode.

Path-like objects for font files

load(), load_path() and save() now accept os.PathLike objects, such as pathlib.Path, in addition to strings.

Other changes

Python 3.15

Pillow 12.3.0 had wheels built against Python 3.15 beta, available as a preview to help projects prepare for 3.15, and to ensure Pillow could be used immediately at the release of 3.15.0 final (2026-10-01, PEP 790).

Pillow 13.0.0 now officially supports Python 3.15.

Faster import

Importing Pillow takes around half the time it did in Pillow 12.3.0. PIL.Image no longer imports logging, modules that are only needed for some operations, such as tempfile, are now imported on first use, and on Python 3.15 Pillow also declares __lazy_modules__ to defer its remaining standard library imports (PEP 810). On an Apple Silicon Mac, from PIL import Image took 24 ms on Python 3.14 and 23 ms on Python 3.15 with Pillow 12.3.0, and now takes 12 ms and 10 ms.

PyPy 3.12

Wheels are now also built for PyPy 3.12, in addition to PyPy 3.11.

Faster JPEG 2000, TIFF, FreeType and LittleCMS in macOS and Linux wheels

Several of the libraries bundled in the macOS and Linux wheels were built without compiler optimisation. The Windows wheels were not affected.

These libraries are now compiled with -O2. Compared to Pillow 12.3.0, the wheels

  • decode JPEG 2000 images 2 to 3 times faster and encode them about twice as fast,

  • read and write LZW-compressed TIFF images about twice as fast,

  • render text with ImageFont up to 3 times faster, and

  • apply ImageCms transforms up to 3 times faster.

PNG, JPEG, WebP and AVIF are unaffected, as those libraries were already built with optimisation.

Performance improvements

Building on the optimizations in Pillow 12.3.0, more of the C code has been restructured so that compilers can vectorize it. Compared to Pillow 12.3.0, offset() is up to 46 times faster, getbbox() up to 24 times faster, flipping and rotating by 180 degrees with transpose() up to 10 times faster, filter() up to 3.5 times faster, difference() up to 3 times faster, and converting P and PA images to RGB and RGBA up to 2.5 times faster. Saving an optimized GIF with a shared palette is around 6 times faster and uses less memory.

Wheels now use HarfBuzz for hinting

FreeType can use HarfBuzz to work out which glyphs a script covers, so that its auto-hinter can derive hinting “Blue Zones” for scripts such as Arabic.

Only the Windows wheels were actually built with this support. The macOS wheels disabled HarfBuzz interoperability outright, and the Linux wheels left it up to FreeType’s configure, which in turn fell back to loading HarfBuzz by its unversioned name at runtime.

Since the bundled copy of HarfBuzz is renamed when the wheel is audited and repaired for distribution, that lookup only succeeded if a separate copy happened to be installed on the system with the expected name (e.g. libharfbuzz.so.0).

All wheels are now linked against the HarfBuzz that Pillow ships, which may slightly change how text is rendered in scripts that rely on the auto-hinter.

Thread safety for free-threaded Python

Critical sections now protect most operations on core image objects, such as conversion, pasting, filtering and pixel access, improving thread safety in the free-threaded build of Python. AVIF decoding also no longer blocks other threads.

Resizing images with alpha

The reducing_gap argument to resize() is now applied to RGBA and LA images. Previously, these modes ignored the argument, including when called by thumbnail(), which uses reducing_gap=2.0 by default.

Preserved transparency when padding or expanding images in ImageOps

pad() and expand() now preserve the transparency value from the source image’s info dictionary. Transparent pixels in L, P and RGB images previously became opaque when a border was added. New palette colors are allocated without using reserved transparency entries.

Fixed reading FITS images with more than 8 bits per pixel

FITS data is big-endian, but Pillow was reading 16-bit and 32-bit integer data, and 32-bit and 64-bit floating point data, in the native byte order. These images now load with the correct pixel values.

Fixed accuracy of 16-bit image filters and conversions

Applying a kernel filter to an I;16 image gave incorrect results for pixel values above 255, and converting an I;16N image to I or F clipped values to 255. Both now preserve the full 16-bit range.

Saving APNG frames with different palettes

An APNG file can only contain a single palette. When saving P mode frames with different palettes, the image is now saved as RGB, or RGBA if any palette contains alpha, instead of writing the later frames with the wrong colors.

Fixed ImageStat variance rounding

Fixed a floating-point rounding error that may have produced a negative variance in var.

Fixed ImageChops.offset() for 16-bit images

offset() now works correctly for 16-bit-per-channel images.

Fixed drawing on I;16B images

When drawing on an I;16B image, some drawing methods wrote in the wrong byte order. All drawing operations now use the byte order of the image mode.

Fixed histograms for masked LA, La and PA images

When histogram() was given a mask for a two-band image (LA, La or PA), the second band was incorrectly derived from the first band of the input image, rather than the alpha band.

Fixed exporting Arrow arrays from images with non-contiguous rows

Exporting an image as an Arrow array assumed that its rows were contiguous. However, this was not the case for

  • images whose rows were padded to an alignment returned incorrect data,

  • empty images,

  • images created by frombuffer(), and

  • images memory-mapped from uncompressed files.

These images crashed instead.

Images with rows that are not contiguous now raise a ValueError, and the others are exported correctly.

Fixed combining GIF frames with missing durations

Saving identical consecutive GIF frames no longer raises a KeyError when an earlier frame has no duration and a later frame does.

PNG palettes are no longer padded when saving with the bits argument

When saving a P mode image as a PNG with the bits argument, Pillow used to pad the PLTE chunk with zeroes up to 2 ** bits entries. The PLTE chunk now only contains as many entries as the image’s palette actually has, matching the behavior already used when bits is not specified.

Restore image position if ImageSequence.all_frames() fails

If processing a frame in all_frames() raises an exception, the given image will now seek back to its original frame.

Clean up TIFF encoders before closing their output files

TIFF saving now explicitly cleans up the libtiff encoder, including when saving raises an exception. Previously, delayed encoder destruction could access a closed file descriptor after it had been reused for another file, disturbing that file’s position or contents.

tobytes() also now explicitly cleans up its encoder on both success and failure, rather than relying on object destruction.

Fixed reading PSD layers at the wrong file offset

When seeking to a layer of a PSD image, the pixel data was read from the wrong position in the file, so seek() returned unrelated bytes reinterpreted as image data instead of that layer’s actual content.

Fixed opening XBM images with comments

XBM images with C-style /* ... */ comments in the header could not be identified. Comments on separate lines before the bits declaration are now skipped.

Clarified FriBiDi shim license

The SBOM published with Pillow 12.3.0 listed the vendored FriBiDi shim (src/thirdparty/fribidi-shim/) as LGPL-2.1-or-later. The shim was written for Pillow and is covered by Pillow’s MIT-CMU license. Its header declares the parts of FriBiDi’s public API needed to load FriBiDi at runtime, and contains no FriBiDi code. FriBiDi itself is not included in Pillow’s wheels.

The shim’s source files now include SPDX license identifiers, and the SBOM has been corrected.

ImagePath.Path.compact() on an empty path

compact() is documented to return the number of points removed. On a path with no points, it instead added a point whose y coordinate was read from uninitialized memory, and returned -1. compact() now leaves an empty path unchanged and returns 0.