psd-tools is a package for reading Adobe Photoshop PSD files (as described in specification) to Python data structures.
pip install psd-tools
There are also optional dependencies:
docopt for command-line interface:
pip install docopt
pip install Pillow
pip install packbits
(Pymaging installation instructions are available in pymaging docs).
In order to extract images from 32bit PSD files PIL/Pillow must be built with LITTLECMS support.
Load an image:
>>> from psd_tools import PSDImage >>> psd = PSDImage.load('my_image.psd')
Read image header:
>>> psd.header PsdHeader(number_of_channels=3, height=200, width=100, depth=8, color_mode=RGB)
Access its layers:
>>> psd.layers [<psd_tools.Group: 'Group 2', layer_count=1>, <psd_tools.Group: 'Group 1', layer_count=1>, <psd_tools.Layer: 'Background', size=100x200, x=0, y=0>]
Work with a layer group:
>>> group2 = psd.layers >>> group2.name Group 2 >>> group2.visible True >>> group2.closed False >>> group2.opacity 255 >>> from psd_tools.constants import BlendMode >>> group2.blend_mode == BlendMode.NORMAL True >>> group2.layers [<psd_tools.Layer: 'Shape 2', size=43x62, x=40, y=72)>]
Work with a layer:
>>> layer = group2.layers >>> layer.name Shape 2 >>> layer.bbox BBox(x1=40, y1=72, x2=83, y2=134) >>> layer.bbox.width, layer.bbox.height (43, 62) >>> layer.visible, layer.opacity, layer.blend_mode (True, 255, u'norm') >>> layer.as_PIL() <PIL.Image.Image image mode=RGBA size=43x62 at ...>
Export a single layer:
>>> layer_image = layer.as_PIL() >>> layer_image.save('layer.png')
Export the merged image:
>>> merged_image = psd.as_PIL() >>> merged_image.save('my_image.png')
The same using Pymaging:
>>> merged_image = psd.as_pymaging() >>> merged_image.save_to_path('my_image.png') >>> layer_image = layer.as_pymaging() >>> layer_image.save_to_path('layer.png')
Export layer group (experimental):
>>> group_image = group2.as_PIL() >>> group_image.save('group.png')
Why yet another PSD reader?
There are existing PSD readers for Python:
- there is a PSD reader in PIL library;
- it is possible to write Python plugins for GIMP.
PSD reader in PIL is incomplete and contributing to PIL is complicated because of the slow release process, but the main issue with PIL for me is that PIL doesn't have an API for layer groups.
GIMP is cool, but it is a huge dependency, its PSD parser is not perfect and it is not easy to use GIMP Python plugin from your code.
I also considered contributing to pypsd or psdparse, but they are GPL and I was not totally satisfied with the interface and the code (they are really fine, that's me having specific style requirements).
So I finally decided to roll out yet another implementation that should be MIT-licensed, systematically based on the specification (it turns out the specs are incomplete and sometimes incorrect though); parser should be implemented as a set of functions; the package should have tests and support both Python 2.x and Python 3.x.
The process of handling a PSD file is split into 3 stages:
- "Reading": the file is read and parsed to low-level data structures that closely match the specification. No user-accessible images are constructed; image resources blocks and additional layer information are extracted but not parsed (they remain just keys with a binary data). The goal is to extract all information from a PSD file.
- "Decoding": image resource blocks and additional layer information blocks are parsed to a more detailed data structures (that are still based on a specification). There are a lot of PSD data types and the library currently doesn't handle them all, but it should be easy to add the parsing code for the missing PSD data structures if needed.
After (1) and (2) we have an in-memory data structure that closely resembles PSD file; it should be fairly complete but very low-level and not easy to use. So there is a third stage:
- "User-facing API": PSD image is converted to an user-friendly object that supports layer groups, exporting data as PIL.Image or pymaging.Image, etc.
Stage separation also means user-facing API may be opinionated: if somebody doesn't like it then it should possible to build an another API based on lower-level decoded PSD file.
psd-tools tries not to throw away information from the original PSD file; even if the library can't parse some info, this info will be likely available somewhere as raw bytes (open a bug if this is not the case). This should make it possible to modify and write PSD files (currently not implemented; contributions are welcome).
- reading of RGB and RGBA images;
- 8bit, 16bit and 32bit channels;
- all PSD compression methods are supported (not only the most common RAW and RLE);
- image ICC profile is taken into account;
- most important (imho) 23 image resource types and 12 tagged block types are decoded;
- there is an optional Cython extension to make the parsing fast;
- very basic & experimental layer merging.
- reading of CMYK, Duotone, LAB, etc. images;
- many image resource types and tagged blocks are not decoded (they are attached to the result as raw bytes);
- this library can't reliably blend layers together: it is possible to export a single layer and to export a final image, but rendering of e.g. layer group may produce incorrect results;
- the decoding of Descriptor structures is very basic;
- the writing of PSD images is not implemented;
- only 8bit images can be converted to pymaging.Image;
- layer merging currently doesn't work with Pymaging.
If you need some of unimplemented features then please fire an issue or implement it yourself (pull requests are welcome in this case).
Development happens at github and bitbucket:
The main issue tracker is at github: https://github.com/kmike/psd-tools/issues
Feel free to submit ideas, bugs, pull requests (git or hg) or regular patches.
In case of bugs it would be helpful to provide a small PSD file demonstrating the issue; this file may be added to a test suite.
Unfortunately I don't have a license for Adobe Photoshop and use GIMP for testing; PNG screenshots may be necessary in cases where GIMP fails.
In order to run tests, install tox and type
from the source checkout.
The license is MIT.