django-imagefield

https://travis-ci.org/matthiask/django-imagefield.svg?branch=master Documentation Status

Heavily based on django-versatileimagefield, but with a few important differences:

  • The amount of code is kept at a minimum. django-versatileimagefield has several times as much code (without tests).
  • Generating images on-demand inside rendering code is made hard on purpose. Instead, images are generated when models are saved and also by running the management command process_imagefields.
  • django-imagefield does not depend on a fast storage or a cache to be and stay fast, at least as long as the image width and height is saved in the database. An important part of this is never determining whether a processed image exists in the hot path at all (except if you force it).
  • django-imagefield fails early when image data is incomplete or not processable by Pillow for some reason.
  • django-imagefield allows adding width, height and PPOI (primary point of interest) fields to the model by adding auto_add_fields=True to the field instead of boringly and verbosingly adding them yourself.

Replacing existing uses of django-versatileimagefield requires the following steps:

  • from imagefield.fields import ImageField as VersatileImageField, PPOIField
  • Specify the image sizes by either providing ImageField(formats=...) or adding the IMAGEFIELD_FORMATS setting. The latter overrides the former if given.
  • Convert template code to access the new properties (e.g. instance.image.square instead of instance.image.crop.200x200 when using the IMAGEFIELD_FORMATS setting below).
  • When using django-imagefield with a PPOI, make sure that the PPOI field is also added to ModelAdmin or InlineModelAdmin fieldsets, otherwise you’ll just see the image, but no PPOI picker. Contrary to django-versatileimagefield the PPOI field is editable itself, which avoids apart from other complexities a pitfall with inline form change detection.
  • Add "imagefield" to INSTALLED_APPS.

If you used e.g. instance.image.crop.200x200 and instance.image.thumbnail.800x500 before, you should add the following setting:

IMAGEFIELD_FORMATS = {
    # image field path, lowercase
    'yourapp.yourmodel.image': {
        'square': ['default', ('crop', (200, 200))],
        'full': ['default', ('thumbnail', (800, 500))],

        # The 'full' spec is equivalent to the following format
        # specification in terms of image file produced (the
        # resulting file name is different though):
        # 'full': [
        #     'autorotate', 'process_jpeg', 'process_gif', 'autorotate',
        #     ('thumbnail', (800, 500)),
        # ],
        # Note that the exact list of default processors may
        # change in the future.
    },
}

After running ./manage.py process_imagefields once you can now use use instance.image.square and instance.image.thumbnail in templates instead. Note that the properties on the image file do by design not check whether thumbs exist.

Image processors

django-imagefield uses an image processing pipeline modelled after Django’s middleware.

The following processors are available out of the box:

  • autorotate: Autorotates an image by reading the EXIF data.
  • process_jpeg: Converts non-RGB images to RGB, activates progressive encoding and sets quality to a higher value of 90.
  • process_gif: Preserves transparency and palette data in resized images.
  • preserve_icc_profile: As the name says.
  • thumbnail: Resizes images to not exceed a bounding box.
  • crop: Crops an image to the given dimensions, also takes the PPOI (primary point of interest) information into account if provided.
  • default: The combination of autorotate, process_jpeg, process_gif and preserve_icc_profile. Additional default processors may be added in the future. It is recommended to use default instead of adding the processors one-by-one.

Processors can be specified either using their name alone, or if they take arguments, using a tuple where the first entry is the processors’ name and the rest are positional arguments.

You can easily register your own processors or even override built-in processors if you want to:

from imagefield.processing import register

# You could also write a class with a __call__ method, but I really
# like the simplicity of functions.

@register
def my_processor(get_image, ...):
    def processor(image, context):
        # read some information from the image...
        # or maybe modify it, but it's mostly recommended to modify
        # the image after calling get_image

        image = get_image(image, context)

        # modify the image, and return it...
        modified_image = ...
        # maybe modify the context...
        return modified_image
    return processor

The processor’s name is taken directly from the registered object.

An example processor which converts images to grayscale would look as follows:

from PIL import ImageOps
from imagefield.processing import register

@register
def grayscale(get_image):
    def processor(image, context):
        image = get_image(image, context)
        return ImageOps.grayscale(image)
    return processor

Now include "grayscale" in the processing spec for the image where you want to use it.

The processing context

The context is a namespace with the following attributes (feel free to add your own):

  • processors: The list of processors.
  • name: The name of the resulting image relative to its storages’ root.
  • extension: The extension of the source and target.
  • ppoi: The primary point of interest as a list of two floats between 0 and 1.
  • save_kwargs: A dictionary of keyword arguments to pass to PIL.Image.save.

The ppoi, extension, processors and name attributes cannot be modified when running processors anymore. Under some circumstances extension and name will not even be there.

If you want to modify the extension or file type, or create a different processing pipeline depending on facts not known when configuring settings you can use a callable instead of the list of processors. The callable will receive the fieldfile and the context instance and must at least set the context’s processors attribute to something sensible. Just as an example here’s an image field which always returns JPEG thumbnails:

from imagefield.processing import register

@register
def force_jpeg(get_image):
    def processor(image, context):
        image = get_image(image, context)
        context.save_kwargs["format"] = "JPEG"
        context.save_kwargs["quality"] = 90
        return image
    return processor

def jpeg_processor_spec(fieldfile, context):
    context.extension = ".jpg"
    context.processors = [
        "force_jpeg",
        "autorotate",
        ("thumbnail", (200, 200)),
    ]

class Model(...):
    image = ImageField(..., formats={"thumb": jpeg_processor_spec})

Of course you can also access the model instance through the field file by way of its fieldfile.instance attribute and use those informations to customize the pipeline.

Development

django-imagefield uses flake8 and black to keep the code clean and formatted. Run both using tox:

tox -e style

The easiest way to build the documentation and run the test suite is also by using tox:

tox -e docs  # Open docs/build/html/index.html
tox -e tests

Change log

`0.8`_ (2019-06-21)

  • BACKWARDS INCOMPATIBLE: Changed processing to pass additional processors’ arguments as positional arguments instead of as a single list. This change only affects custom processors, no changes are necessary for users of the library, except if for example you passed arguments to processors such as default, autorotate etc.
  • Fixed a test to assume less about the error message for corrupt images.
  • Localize the corrupt image validation errors.
  • Stopped calling the storage’s delete() method for non-existing images.
  • Made the field resilient against NULL values from the database.

0.7 (2018-10-18)

  • Made error reporting in process_imagefields include more info.
  • Made image field validation catch errors while determining the image dimension too.
  • Fixed a problem where older versions of Django didn’t allow specifying the chunk size for iterating over querysets.
  • Modified django-imagefield’s internals to allow changing the type and extension of generated images by way of dynamically specifying the processing pipeline.
  • Changed the API of the get_image callable in processors to only return the image without the context (since the context is mutable and available already).

0.6 (2018-09-13)

  • Fixed a crash where unpickling image fields would fail.
  • Changed process_imagefields to skip exclude model instances with an empty image field.
  • Changed the thumbnail processor to not upscale images.
  • Made process_imagefields not load the whole queryset at once to avoid massive slowdowns while determining the width and height of images (if those fields aren’t filled in yet).
  • Added housekeeping options to process_imagefields. The only method implemented right now is --housekeep blank-on-failure which empties image fields where processing fails.
  • Changed process_imagefields to process items in a deterministic order.
  • Clarified the processors spec documentation a bit and added an example how to write a processor of your own.

0.5 (2018-08-15)

  • Dropped support for using image fields without associated height and width fields, because it is almost (?) always a really bad idea performance-wise.
  • Fixed a bug where processed image names on Python 2 were different than those generated using Python 3. This bug affects only installations still using Python 2. Rerun ./manage.py process_imagefields --all after upgrading.

0.4 (2018-08-13)

  • Added compatibility with Django 1.8 for prehistoric projects.
  • Polished tests and docs a bit.

0.3 (2018-05-29)

  • BACKWARDS INCOMPATIBLE: Changed the filename generation method to preserve the filename part of the original file for SEO purposes etc. You should run ./manage.py process_imagefields --all, and optionally empty the __processed__ folder before doing that if you do not want to keep old images around.
  • Improved progress reporting in process_imagefields.
  • Added a call to instance.save() in process_imagefields so that width and height fields are saved (if any).
  • Added accept="image/*" attribute to the file upload widget.
  • Replaced the full image in the admin widget with an ad-hoc thumbnail.
  • Fixed a bug where blank imagefields would not work correctly in the administration interface.
  • Switched the preferred quote to " and started using black to automatically format Python code.

0.2 (2018-03-28)

  • Rename management command to process_imagefields, and add --all option to process all imagefields.
  • Fixed a bug where not all image fields from base classes were picked up for processing by process_imagefields.
  • Added the IMAGEFIELD_AUTOGENERATE setting, which can be set to a list of image fields (in app.model.field notation, lowercased) to only activate automatic processing of images upon model creation and update for a few specific fields, or to False to disable this functionality for all fields.
  • Added system checks which warn when width_field and height_field are not used.
  • Changed process_imagefields to process image fields in alphabetic order. Also, made cosmetic changes to the progress output.
  • Added a test which verifies that generating processed image URLs is not slowed down by potentially slow storages (e.g. cloud storage)
  • Fixed the PPOI JavaScript to not crash when some imagefields have no corresponding PPOI input.

0.1 (2018-03-27)

  • First release that should be fit for public consumption.