./configure make make install
This document outlines the interfaces to GIMP-Python, which is a set of Python modules that act as a wrapper to libgimp allowing the writing of plug-ins for GIMP. In this way, GIMP-Python is similar to Script-Fu, except that you can use the full set of Python extension modules from the plug-in.
GIMP-Python is a scripting extension for GIMP, similar to Script-Fu. The main difference is in what is called first. In Script-Fu, the script-fu plug-in executes the script, while in GIMP-Python the script is in control.
In fact, you will find that the GIMP-Python scripts start with the line !/usr/bin/python . The GIMP extension is loaded with the familiar import command.
Another point of difference between GIMP-Python and Script-Fu is that GIMP-Python stores images, layers, channels and other types as objects rather than just storing their ID. This allows better type checking that is missing from Script-Fu, and allows those types to act as objects, complete with attributes and methods.
Also, GIMP-Python is not limited to just calling procedures from the PDB. It also implements the rest of libgimp , including tiles and pixel regions, and access to other lower level functions.
GIMP-python consists of a Python module written in C and some native python support modules. You can build PyGIMP with the commands:
./configure make make install
This will build and install gimpmodule and its supporting modules, and install the sample plug-ins in GIMP's plug-in directory.
The majority of code in this package resides in gimpmodule.c , but this provides a poor interface for implementing some portions of a plugin. For this reason, there is a python module called plugin.py that sets out a structure for plug-ins and implements some things that were either too difficult or impossible to do in C.
The main purpose of plugin.py was to implement an object oriented structure for plug-ins. As well as this, it handles tracebacks, which are otherwise ignored by libgimp , and gives a method to call other GIMP-Python plug-ins without going through the procedural database.
As in a lot of manuals, the first thing you examine is an example, so here is an example. I have included it before explaining what it does to allow more advanced programmers to see the structure up front. It is a translation of the clothify Script-Fu extension:
#!/usr/bin/env python import math from gimpfu import * def python_clothify(timg, tdrawable, bx=9, by=9, azimuth=135, elevation=45, depth=3): width = tdrawable.width height = tdrawable.height img = gimp.Image(width, height, RGB) img.disable_undo() layer_one = gimp.Layer(img, "X Dots", width, height, RGB_IMAGE, 100, NORMAL_MODE) img.add_layer(layer_one, 0) pdb.gimp_edit_fill(layer_one, BACKGROUND_FILL) pdb.plug_in_noisify(img, layer_one, 0, 0.7, 0.7, 0.7, 0.7) layer_two = layer_one.copy() layer_two.mode = MULTIPLY_MODE layer_two.name = "Y Dots" img.add_layer(layer_two, 0) pdb.plug_in_gauss_rle(img, layer_one, bx, 1, 0) pdb.plug_in_gauss_rle(img, layer_two, by, 0, 1) img.flatten() bump_layer = img.active_layer pdb.plug_in_c_astretch(img, bump_layer) pdb.plug_in_noisify(img, bump_layer, 0, 0.2, 0.2, 0.2, 0.2) pdb.plug_in_bump_map(img, tdrawable, bump_layer, azimuth, elevation, depth, 0, 0, 0, 0, True, False, 0) gimp.delete(img) register( "python_fu_clothify", "Make the specified layer look like it is printed on cloth", "Make the specified layer look like it is printed on cloth", "James Henstridge", "James Henstridge", "1997-1999", "<Image>/Filters/Artistic/_Clothify...", "RGB*, GRAY*", [ (PF_INT, "x_blur", "X blur", 9), (PF_INT, "y_blur", "Y blur", 9), (PF_INT, "azimuth", "Azimuth", 135), (PF_INT, "elevation", "Elevation", 45), (PF_INT, "depth", "Depth", 3) ], , python_clothify) main()
In this plugin, a number of modules are imported. The important ones are:
gimpfu : this module provides a simple interface for writing plug-ins, similar to what script-fu provides. It provides the GUI for entering in parameters in interactive mode and performs some sanity checks when registering the plugin.
By using "from gimpfu import *", this module also provides an easy way to get all the commonly used symbols into the plug-in's namespace.
gimp : the main part of the gimp extension. This is imported with gimpfu.
gimpenums : a number of useful constants. This is also automatically imported with gimpfu.
The pdb variable is a variable for accessing the procedural database. It is imported into the plug-in's namespace with gimpfu for convenience.
With pygimp-0.4, the gimpfu module was introduced. It simplifies writing plug-ins a lot. It handles the run mode (interactive, non interactive or run with last values), providing a GUI for interactive mode and saving the last used settings.
Using the gimpfu plugin, all you need to do is write the function that should be run, make a call to register, and finally a call to main to get the plugin started.
If the plugin is to be run on an image, the first parameter to the plugin function should be the image, and the second should be the current drawable (do not worry about the run_mode parameter). Plug-ins that do not act on an existing image (and hence go in the toolbox's menus) do not need these parameters. Any other parameters are specific to the plugin.
After defining the plugin function, you need to call register to register the plugin with gimp (When the plugin is run to query it, this information is passed to gimp. When it is run interactively, this information is used to construct the GUI). The parameters to register are:
Most of these parameters are quite self explanatory. The menupath option should start with <Image>;/ for image plug-ins and <Toolbox>/ for toolbox plug-ins. The remainder of the menupath is a slash separated path to its menu item.
The params parameter holds a list parameters for the function. It is a list of tuples. Note that you do not have to specify the run_type, image or drawable parameters, as gimpfu will add these automatically for you. The tuple format is (type, name, description, default [, extra]). The allowed type codes are:
These values map onto the standard PARAM_* constants. The reason to use the extra constants is that they give gimpfu more information, so it can produce a better interface (for instance, the PF_FONT type is equivalent to PARAM_STRING, but in the GUI you get a small button that will bring up a font selection dialog).
The PF_SLIDER, PF_SPINNER and PF_ADJUSTMENT types require the extra parameter. It is of the form (min, max, step), and gives the limits for the spin button or slider.
The results parameter is a list of 3-tuples of the form (type, name, description). It defines the return values for the function. If there is only a single return value, the plugin function should return just that value. If there is more than one, the plugin function should return a tuple of results.
The final parameter to register is the plugin function itself.
After registering one or more plugin functions, you must call the main function. This will cause the plugin to start running. A GUI will be displayed when needed, and your plugin function will be called at the appropriate times.
The procedural database is a registry of things gimp and its plug-ins can do. When you install a procedure for your plugin, you are extending the procedural database.
The procedural database is self documenting, in that when you install a procedure in it, you also add documentation for it, its parameters and return values.
In Gimp-Python, the procedural database is represented by the object gimp.pdb. In most of my plug-ins, I make an assignment from gimp.pdb to pdb for convenience.
You can query the procedural database with pdb's method query. Its specification is:
pdb.query(name, [blurb, [help, [author, [copyright, [date, [type]]]]]])
Each parameter is a regular expression that is checked against the corresponding field in the procedural database. The method returns a list of the names of matching procedures. If query is called without any arguments, it will return every procedure in the database.
Procedures can be accessed as procedures, or by treating pdb as a mapping object. As an example, the procedure gimp_edit_fill can be accessed as either pdb.gimp_edit_fill or pdb['gimp_edit_fill']. The second form is mainly for procedures whose names are not valid Python names (eg in script-fu-…, the dashes are interpreted as minuses).
These procedure objects have a number of attribute:
The name of the procedure.
A short piece of information about the procedure.
More detailed information about the procedure.
The author of the procedure.
The copyright holder for the procedure (usually the same as the author).
The date when the procedure was written.
The type of procedure. This will be one of PROC_PLUG_IN, PROC_EXTENSION or PROC_TEMPORARY.
The number of parameters the procedure takes.
The number of return values the procedure gives.
A description of parameters of the procedure. It takes the form of a tuple of 3-tuples, where each 3-tuple describes a parameter. The items in the 3-tuple are a parameter type (one of the PARAM_* constants), a name for the parameter, and a description of the parameter.
A description of the return values. It takes the same form as the params attribute.
A procedure object may also be called. At this point, Gimp-Python doesn't support keyword arguments for PDB procedures. Arguments are passed to the procedure in the normal method. The return depends on the number of return values:
If there are zero return values, None is returned.
If there is only a single return value, it is returned.
If there are more return values, then they are returned as a tuple.
For more information on invoking PDB procedures, please see the example plug-ins. For information on individual procedures, please see the PDB Browser plugin (in the Xtns menu). It allows you to peruse to the database interactively.
The gimp module contains a number of procedures and functions, as well as the definitions of many gimp types such as images, and the procedural database. This section explains the base level procedures.
There are a number of functions in the gimp module that are used to create the objects used to make up an image in GIMP. Here is a set of descriptions of these constructors:
This procedure creates an image with the given dimensions and type (type is one of RGB , GRAY or INDEXED ).
Create a new layer called name, with the given dimensions and type (one of the *_IMAGE constants), opacity (float between 0 and 100) and a mode (one of the *_MODE constants). The layer can then be added to the image with the img.add_layer method.
Create a new channel object with the given dimensions, opacity and colour (one of the *_CHANNEL constants). This channel can then be added to an image.
Create a new display window for the given image. The window will not be displayed until a call to gimp.displays_flush is made.
Create a new parasite. The parasite can then be attached to gimp, an image or a drawable. This is only available in gimp >= 1.1
When any of these objects get removed from memory (such as when their name goes out of range), the gimp thing it represents does not get deleted with it (otherwise when your plugin finished running, it would delete all its work). In order to delete the thing the Python object represents, you should use the gimp.delete procedure. It deletes the gimp thing associated with the Python object given as a parameter. If the object is not an image, layer, channel, drawable or display gimp.delete does nothing.
There are a number of functions that can be used to gather information about the environment the plugin is running in:
Returns the current colour cube.
Returns the current gamma correction.
Returns non-zero if a colour map has been installed.
Returns non-zero if GIMP is using X shared memory.
Returns the file name of the GTK configuration file.
These functions alter the currently selected foreground and background.
Returns a 3-tuple containing the current background colour in RGB form.
Returns a 3-tuple containing the current foreground colour in RGB form.
Sets the current background colour. The three arguments can be replaced by a single 3-tuple like that returned by gimp.get_background.
Sets the current foreground colour. Like gimp.set_background, the arguments may be replaced by a 3-tuple.
These functions perform operations on gradients:
Returns the name of the active gradient.
Sets the active gradient.
Returns a list of the names of the available gradients.
Returns a list of num samples, where samples consist of 4-tuples of floats representing the red, green, blue and alpha values for the sample.
Similar to gimp.gradients_sample_uniform, except the samples are taken at the positions given in the list of floats pos instead of uniformly through the gradient.
These functions either install procedures into the PDB or alert gimp to their special use (eg as file handlers).
For simple plug-ins, you will usually only need to use register from gimpfu.
This procedure is used to install a procedure into the PDB. The first eight parameters are strings, type is a one of the PROC_* constants, and the last two parameters are sequences describing the parameters and return values. Their format is the same as the param and ret_vals methods or PDB procedures.
This procedure is used to install a procedure into the PDB temporarily. That is, it must be added again every time gimp is run. This procedure will be called the same way as all other procedures for a plugin.
This removes a temporary procedure from the PDB.
This procedure tells GIMP that the PDB procedure name can load files with extensions and prefixes (eg http:) with magic information magics.
This procedure tells GIMP that the PDB procedure name can load files with extensions and prefixes (eg http:).
This procedure tells GIMP that the PDB procedure name can save files with extensions and prefixes (eg http:).
These are the other functions in the gimp module.
This function is the one that controls the execution of a Gimp- Python plugin. It is better to not use this directly but rather subclass the plugin class, defined in the gimpplugin Module.
The procedural database object.
(Re)Initialise the progress meter with label (or the plugin name) as a label in the window.
Set the progress meter to percent done.
Returns a list of all the image objects.
Stops execution immediately and exits.
Update all the display windows.
The maximum width of a tile.
The maximum height of a tile.
Set the size of the tile cache in kilobytes.
Set the size of the tile cache in tiles.
Get the information associated with key. The data will be a string. This function should probably be used through the gimpshelf Module.
Set the information in the string data with key. The data will persist for the whole gimp session. Rather than directly accessing this function, it is better to go through the gimpshelf Module.
Tells gimp that the plugin has finished its work, while keeping the plugin connection open. This is used by an extension plugin to tell gimp it can continue, while leaving the plugin connection open. This is what the script-fu plugin does so that only one scheme interpreter is needed.
Makes the plugin check for messages from gimp. generally this is not needed, as messages are checked during most calls in the gimp module.
In gimp >= 1.1, it is possible to attach arbitrary data to an image through the use of parasites. Parasites are simply wrappers for the data, containing its name and some flags. Parasites have the following parameters:
The data for the parasite — a string
The flags for the parasite
True if this parasite is persistent
True if this parasite is undoable
The name of the parasite
Parasites also have the methods copy, is_type and has_flag.
There is a set of four functions that are used to manipulate parasites. They exist as functions in the gimp module, and methods for image and drawable objects. They are:
find a parasite by its name.
Attach a parasite to this object.
Create a new parasite and attach it.
Detach the named parasite
Gimp-Python implements a number of special object types that represent the different types of parameters you can pass to a PDB procedure. Rather than just making these place holders, I have added a number of members and methods to them that allow a lot of configurability without directly calling PDB procedures.
There are also a couple of extra objects that allow low level manipulation of images. These are tile objects (working) and pixel regions (not quite finished).
This is the object that represents an open image. In this section, image represents a generic image object.
This is the active channel of the image. You can also assign to this member, or None if there is no active channel.
This is the active layer of the image. You can also assign to this member, or None if there is no active layer.
This is the type of the image (eg RGB, INDEXED).
This is a list of the channels of the image. Altering this list has no effect, and you can not assign to this member.
This is the colour map for the image.
This is the filename for the image. A file load or save handler might assign to this.
This is the height of the image. You can't assign to this member.
The floating selection layer, or None if there is no floating selection.
This is a list of the layers of the image.
The selection mask for the image.
This is the width of the image. You can't assign to this member.
Adds channel to image in position position.
Adds layer to image in position position.
Adds the mask mask to layer.
Unsets the dirty flag on the image.
Disables undo for image.
Enables undo for image. You might use these commands round a plugin, so that the plug-in's actions can be undone in a single step.
Returns the resulting layer after merging all the visible layers, discarding non visible ones and stripping the alpha channel.
Returns true if component (one of the *_CHANNEL constants) is active.
Returns true if component is visible.
Sets the activeness of component.
Sets the visibility of component.
Merges the visible layers of image using the given merge type.
Returns the layer that is visible at the point (x,y), or None if no layer matches.
Removes channel from image.
Removes layer from image.
Removes the mask from layer, with the given mode (either APPLY or DISCARD).
Resizes the image to size (width, height) and places the old contents at position (x,y).
These objects represent a GIMP Image's colour channels. In this section, channel will refer to a generic channel object.
The colour of the channel.
The height of the channel.
The width of the channel.
The image the channel belongs to, or None if it isn't attached yet.
The channel's layer (??) or None if one doesn't exist.
Non zero if the channel is a layer mask.
The name of the channel.
The opacity of the channel.
The show_masked value of the channel.
Non-zero if the channel is visible.
returns a copy of the channel.
Layer objects represent the layers of a GIMP image. In this section I will refer to a generic layer called layer.
The apply mask setting. (non zero if the layer mask is being composited with the layer's alpha channel).
The number of bytes per pixel.
The edit mask setting. (non zero if the mask is active, rather than the layer).
The height of the layer.
The image the layer is part of, or None if the layer isn't attached.
Non zero if this layer is the image's floating selection.
The layer's mask, or None if it doesn't have one.
The mode of the layer.
The name of the layer.
The opacity of the layer.
The layer's preserve transparency setting.
Adds an alpha component to the layer.
Creates a copy of the layer, optionally with an alpha layer.
Creates a layer mask of type type.
Resizes the layer to (w, h), positioning the original contents at (x,y).
Scales the layer to (w, h), using the specified origin (local or image).
Sets the offset of the layer, relative to the image's origin
Moves the layer to (x, y) relative to its current position.
Both layers and channels are drawables. Hence there are a number of operations that can be performed on both objects. They also have some common attributes and methods. In the description of these attributes, I will refer to a generic drawable called drawable.
The number of bytes per pixel.
Non zero if the drawable is colour.
Non zero if the drawable is greyscale.
Non zero if the drawable has an alpha channel.
The height of the drawable.
The image the drawable belongs to.
Non zero if the drawable uses an indexed colour scheme.
The bounds of the drawable's selection.
The name of the drawable.
The offset of the top left hand corner of the drawable.
The type of the drawable.
Non zero if the drawable is visible.
The width of the drawable.
Fills the drawable with given fill_type (one of the *_FILL constants).
Flush the changes to the drawable.
Creates a pixel region for the drawable. It will cover the region with origin (x,y) and dimensions w x h. The dirty argument sets whether any changes to the pixel region will be reflected in the drawable (default is TRUE). The shadow argument sets whether the pixel region acts on the shadow tiles or not (default is FALSE). If you draw on the shadow tiles, you must call drawable.merge_shadow() for changes to take effect.
Get a tile at (row, col). Either on or off the shadow buffer.
Get the tile that contains the pixel (x, y).
Merge the shadow buffer back into the drawable.
Update the given portion of the drawable.
Tile objects represent the way GIMP stores information. A tile is basically just a 64x64 pixel region of the drawable. The reason GIMP breaks the image into small pieces like this is so that the whole image doesn't have to be loaded into memory in order to alter one part of it. This becomes important with larger images.
In Gimp-Python, you would use Tiles if you wanted to perform some low level operation on the image, instead of using procedures in the PDB. This type of object gives a Gimp-Python plugin the power of a C plugin, rather than just the power of a Script-Fu script. Tile objects are created with either the drawable.get_tile() or drawable.get_tile2() functions. In this section, I will refer to a generic tile object named tile.
All tile members are read only.
The number of bytes per pixel.
If there have been changes to the tile since it was last flushed.
The drawable that the tile is from.
The actual height of the tile.
The actual width of the tile.
The reference count of the tile. (this is independent of the Python object reference count).
Non zero if the tile is part of the shadow buffer.
Flush any changes in the tile. Note that the tile is automatically flushed when the Python object is deleted from memory.
Tile objects also act as a mapping, or sequence. You can access the pixels in the tile in one of two ways. You can either access them with a single number, which refers to its position in the tile (eg. tile  refers to the first pixel in the second row of a 64x64 pixel tile). The other way is with a tuple, representing the coordinates on the tile (eg. tile [0, 1] refers to the first pixel on the second row of the tile).
The type of these subscripts is a string of length tile.bpp. When you assign to a subscript, the dirty flag is automatically set on the tile, so you don't have to explicitly set the flag, or flush the tile.
Pixel region objects give an interface for low level operations to act on large regions of an image, instead of on small 64x64 pixel tiles. In this section I will refer to a generic pixel region called pr. For an example of a pixel region's use, please see the example plugin whirlpinch.py .
The drawable this pixel region is for.
The number of bytes per pixel for the drawable.
The rowstride for the pixel region.
The x coordinate of the top left hand corner.
The y coordinate of the top left hand corner.
The width of the pixel region.
The height of the pixel region.
Non zero if changes to the pixel region will be reflected in the drawable.
Non zero if the pixel region acts on the shadow tiles of the drawable.
resize the pixel region so that it operates on the the region with corner (x, y) with dimensions w x h.
The pixel region acts as a mapping. The index is a 2-tuple with components that are either integers or slices. The subscripts may be read and assigned to. The type of the subscripts is a string containing the binary data of the requested region. Here is a description of the possible operations:
Get/Set the pixel at (x,y)
Get/Set the row starting at (x1, y), width x2 - x1.
Get/Set the column starting at (x, y1), height y2 - y1.
Get/Set the rectangle starting at (x1, y1), width x2 - x1 and height y2 - y1.
This section describes the modules that help make using the gimp module easier. These range from a set of constants to storing persistent data.
This module contains all the constants found in the header libgimp/gimpenums.h , as well as some extra constants that are available in Script-Fu.
This module was fully described in an earlier section. It provides an easy interface for writing plug-ins, where you do not need to worry about run_modes, GUI's and saving previous values. It is the recommended module for writing plug-ins.
This module provides the framework for writing GIMP plug-ins in Python. It gives more flexibility for writing plug-ins than the gimpfu module, but does not offer as many features (such as automatic GUI building).
To use this framework you subclass gimpplugin.plugin like so:
import gimpplugin class myplugin(gimpplugin.plugin): def init(self): # initialisation routines # called when gimp starts. def quit(self): # clean up routines # called when gimp exits (normally). def query(self): # called to find what functionality the plugin provides. gimp.install_procedure("procname", ...) # note that this method name matches the first arg of # gimp.install_procedure def procname(self, arg1, ...): # do what ever this plugin should do
This module gives a nicer interface to the persistent storage interface for GIMP plug-ins. Due to the complicated nature of Python objects (there is often a lot of connections between them), it can be difficult to work out what to store in GIMP's persistent storage. The python interface only allows storage of strings, so this module wraps pickle and unpickle to allow persistent storage of any python object.
Here is some examples of using this module:
>>> from gimpshelf import shelf >>> shelf['james'] = ['forty-two', (42, 42L, 42.0)] >>> shelf.has_key('james') 1 >>> shelf['james'] ['forty-two', (42, 42L, 42.0)]
Anything you store with gimpshelf.shelf will exist until GIMP exits. This makes this interface perfect for when a plugin is executed with the run mode RUN_WITH_LAST_VALS .
As all other plug-ins, the ones written with GIMP Python are registered in the procedural database (See Procedural Database). As such, they can be easily interfaced using Script-Fu; let's consider this GIMP Python plug-in:
#! /usr/bin/env python from gimpfu import * def echo(*args): """Print the arguments on standard output""" print "echo:", args register( "console_echo", "", "", "", "", "", "<Toolbox>/Xtns/Languages/Python-Fu/Test/_Console Echo", "", [ (PF_STRING, "arg0", "argument 0", "test string"), (PF_INT, "arg1", "argument 1", 100 ), (PF_FLOAT, "arg2", "argument 2", 1.2 ), (PF_COLOR, "arg3", "argument 3", (0, 0, 0) ), ], , echo ) main()
Using Script-Fu, one could easily invoke the correct Scheme function call: (python-fu-console-echo RUN-NONINTERACTIVE "another string" 777 3.1416 '(1 0 0)) . There are a couple of details worth mentioning:
The registered procedure name (first parameter of the register() function) is mangled: all underscores are converted to hyphens, to better match the usual Scheme syntatic style (here, console_echo becomes console-echo ). Moreover, a python-fu- prefix is automatically added; it is better not to explicitly add it ourselves, as it will makes the often useful (plug-in-script-fu- eval …) evaluation fail.
The mandatory first parameter to any pdb call (Script-Fu constant RUN-INTERACTIVE , or RUN-NONINTERACTIVE ) discriminates between user-driven or scripted calls. In the case of gimpfu-based plug-ins, it is automatically taken care of internally — interactive calls will dismiss all the remaining arguments, and an interface will be presented (when possible) to the user. The plug-in core Python function ( echo() in this case), never has to deal with it.
Script-Fu is able to process strings, integer and floats literals, and pass them as corresponding first-class objects to Python. It can also pass compound arguments such as colors as tuples in Python by expressing them as Scheme lists (in the (python-fu-console-echo …) call above, fourth argument to the pythonic echo() function will be a 3-tuple of integers representing a pure red color).
All other special purposes gimpfu parameter types (PF_FILE , etc.) can be constructed using those simple literals and list constructs, and received as appropriate objects in Python. If you need booleans, pass them as integers.
All this means that you could easily invoke a GIMP Python plug-in such as the one above directly from your shell using the (plug-in-script- fu-eval …) evaluator:
gimp --no-interface --batch '(python-fu-console-echo RUN-NONINTERACTIVE "another string" 777 3.1416 (list 1 0 0))' '(gimp-quit 1)'
The invocation here was done without an interface since this specific procedure didn't need any.
See the GIMP Script-Fu Documentation to learn more about it.
This package is not yet complete, but it has enough in it to be useful for writing plug-ins for GIMP. If you write any plug-ins that might be useful as examples, please mail me at firstname.lastname@example.org.