8.3.0 (2021-07-01)


CVE-2021-34552: Fix buffer overflow

PIL since 1.1.4 and Pillow since 1.0 allowed parameters passed into a convert function to trigger buffer overflow in Convert.c.

Parsing XML

Pillow previously parsed XMP data using Python’s xml module. However, this module is not secure.

  • getexif() has used xml to potentially retrieve orientation data since Pillow 7.2.0. It has been refactored to use re instead.

  • getxmp() was added to JpegImageFile in Pillow 8.2.0. It will now use defusedxml instead. If the dependency is not present, an empty dictionary will be returned and a warning raised.



JPEG quantization is now automatically converted, but still returned as a dictionary. The convert_dict_qtables method no longer performs any operations on the data given to it, has been deprecated and will be removed in Pillow 10.0.0 (2023-07-01).

API Changes

Changed WebP default “method” value when saving

Previously, it was 0, for the best speed. The default has now been changed to 4, to match WebP’s default, for higher quality with still some speed optimisation.

Default resampling filter for special image modes

Pillow 7.0 changed the default resampling filter to Image.BICUBIC. However, as this is not supported yet for images with a custom number of bits, the default filter for those modes has been reverted to Image.NEAREST.

ImageMorph incorrect mode errors

For apply(), match() and get_on_pixels(), if the image mode is not L, an Exception was thrown. This has now been changed to a ValueError.


XMP data can now be returned for PNG and TIFF images, through getxmp() for each format.

The returned dictionary will start from the base of the XML, meaning that the top level should contain an “xmpmeta” key. JPEG’s getxmp() method has also been updated to this structure.

TIFF getexif()

TIFF tag_v2 data can now be accessed through getexif(). This also provides access to the GPS and EXIF IFDs, through im.getexif().get_ifd(0x8825) and im.getexif().get_ifd(0x8769) respectively.

API Additions


Returns a resized version of the image, set to the maximum width and height within size, while maintaining the original aspect ratio.

To compare it to other ImageOps methods:

  • fit() expands an image until is fills size, cropping the parts of the image that do not fit.

  • pad() expands an image to fill size, without cropping, but instead filling the extra space with color.

  • contain() is similar to pad(), but it does not fill the extra space. Instead, the original aspect ratio is maintained. So unlike the other two methods, it is not guaranteed to return an image of size.

ICO saving: bitmap_format argument

By default, Pillow saves ICO files in the PNG format. They can now also be saved in BMP format, through the new bitmap_format argument:

im.save("out.ico", bitmap_format="bmp")

Other Changes

Added DDS BC5 reading and uncompressed saving

Support has been added to read the BC5 format of DDS images, whether UNORM, SNORM or TYPELESS.

Support has also been added to write the uncompressed format of DDS images.