Table of Contents
GEGL (Generic Graphics Library) is a graph based image processing
GEGL provides infrastructure to do demand based cached non destructive
image editing on larger than RAM buffers. Through babl it provides
support for a wide range of color models and pixel storage formats for
input and output.
• Floating point handling and processing and output of larger 8bit,
16bit integer and 32bit floating point per component buffers larger
• C, vala, C#, Python and Ruby interfaces using a consistent DOM like
graph API to manage processing graphs.
□ Iterative chunk-wise processing.
□ Processes subregions and dependencies.
□ Subgraph caches to aid performance of non-destructive editing.
□ Experimental OpenCL acceleration, with possibility of hybrid
□ Storage of all babl supported formats.
□ Tiled sparse buffers (larger than RAM images).
□ linear buffers (allocated internally or from external
□ On demand tiled mipmapping.
□ inter process shared storage
□ External tile-backends (allow wrapping other tiled buffer
systems to use them through the GeglBuffer API).
□ PNG, JPEG, SVG, EXR, RAW, ffmpeg, v4l and other image sources.
□ Pattern renderers
□ Arithmetic operations
□ link_operations.html#porter_duff[porter duff compositing]
□ SVG filter modes and full set of compositing ops from SVG-1.2
□ Gaussian blur, bilateral-filter, symmetric nearest neighbour,
□ blur, unsharp mask, pixelize and more.
□ Color correction.
□ Text rendering using cairo and pango.
□ HDR exposure merging and tone mapping operations.
□ Most operations operate in scRGB (using 32bit floating point/
HDR linear light RGBA)
• Bounding box based hit detection.
• XML serialization format (not-finalized)
This website is built at the time of the previous GEGL tarball release,
for information about what might change on the way to the next release
follow the following news sources:
For day to day fixes, contributions and changes.
The NEWS file for a list of major new features (also contains older
for known and tracked issues with GEGL and perhaps see the
The mailinglist archives for some discussion and announcement.
For examples of what GEGL’s rendering engine currently can do look at
The GEGL project uses GNOME Bugzilla, a bug-tracking system that allows
us to coordinate bug reports. Bugzilla is also used for enhancement
requests and the preferred way to submit patches for GEGL is to open a
bug report and attach the patch to it. Bugzilla is also the closest you
will find to a roadmap for GEGL.
Below is a list of links to get you started with Bugzilla:
• List of Open Bugs
• List of Open Bugs (excluding enhancement requests)
• List of Enhancement Proposals
• Bugzilla Weekly Summary
You can subscribe to gegl-developer and view the archives here. The
GEGL developer list is the appopriate place to ask development
questions, and get more information about GEGL development in general.
You can email this list at firstname.lastname@example.org.
GEGL development is also discussed in #gegl on GIMPnet (irc.gimp.org).
GEGL is free software; you can redistribute it and/or modify it under
the terms of the GNU Lesser General Public License and GNU General
Public License as published by the Free Software Foundation; either
version 3 of the Licenses, or (at your option) any later version. The
library itself is licensed under LGPL while the sample commandline
application and GUI binary gegl is licensed under GPL.
Many people have contributed to GEGL over time the following lists are
are ordered chronologically according to when they are mentioned in the
Calvin Williamson, Caroline Dahloff, Manish Singh, Jay Cox Daniel
Rogers, Sven Neumann, Michael Natterer, Øyvind Kolås, Philip Lafleur,
Dominik Ernst, Richard Kralovic, Kevin Cozens, Victor Bogado, Martin
Nordholts, Geert Jordaens, Michael Schumacher, John Marshall, Étienne
Bersac, Mark Probst, Håkon Hitland, Tor Lillqvist, Hans Breuer, Deji
Akingunola and Bradley Broom, Hans Petter Jansson, Jan Heller,
email@example.com, Sven Anders, Hubert Figuière, Sam Hocevar, yahvuu
at gmail.com, Nicolas Robidoux, Ruben Vermeersch, Gary V. Vaughan,
James Legg, Henrik Åkesson, Fryderyk Dziarmagowski, Ozan Caglayan,
Tobias Mueller, Nils Philippsen, Adam Turcotte, Danny Robson, Javier
Jardón and Yakkov Selkowitz, Kaja Liiv, Eric Daoust, Damien de Lemeny,
Fabian Groffen, Vincent Untz, Debarshi Ray, Stuart Axon, Kao, Barak
Itkin, Michael Muré, Mikael Magnusson, Patrick Horgan, Tobias
Ellinghaus, Rasmus Hahn, Chantal Racette, John Cupitt, Anthony Thyssen,
Emilio Pozuelo Monfort, Robert Sasu, Massimo Valentini, Hans Lo,
Zbigniew Chyla, David Evans, Javier Jardón, Matteo F. Vescovi and Jan
Garry R. Osgood, Øyvind Kolås, Kevin Cozens and Shlomi Fish.
Jakub Steiner, Øyvind Kolås, Tonda Tavalec
Building from source
GEGL and it’s dependencies are known to work on Linux based systems,
windows with msys/mingw, and probably other platforms.
The latest development snapshot, and eventually stable versions of GEGL
are available at ftp://ftp.gimp.org/pub/gegl/.
The current code under development can be browsed online and cloned
from GNOME git using:
$ git clone git://git.gnome.org/babl
$ git clone git://git.gnome.org/gegl
GEGL currently builds and works on linux, win32 and OSX most probably
also on other systems where glib and gtk+ work.
□ glib (including gobject, and gmodule) 2.16 or newer, which
provides inheritance, dynamic modules, common algorithms and
data structures for C programming.
□ babl 0.0.22 or newer (for pixel-format agnostisism).
□ libpng (png load/export ops, and image magick fallback import)
□ ruby (only if building from git)
• GUI (sandbox for testing ops and the API)
• Optional dependencies for operations.
□ SDL (display op)
□ libjpeg (jpg loader op)
□ libopenexr (exr loader op)
□ libopenraw (raw loader op)
□ cairo, pango (text source op)
□ avcodec, avformat (ff-load and experimental ff-save)
□ librsvg (svg loader)
• Documentation (this document)
To build GEGL type the following in the toplevel source directory:
$ ./configure # or: ./autogen.sh if building from git
$ sudo make install
With GEGL you chain together image processing operations represented by
nodes into a graph. GEGL provides such operations for loading and
storing images, adjusting colors, filtering in different ways,
transforming and compositing images.
GEGL’s programmer/user interface is a Directed Acyclic Graph of nodes.
The DAG expresses a processing chain of operations. A DAG, or any node
in it, expresses a composited and processed image. It is possible to
request rectangular regions in a wide range of pixel formats from any
This howto describes good-to-know things for developing gegl
The public API reference documents the API used for creating things
with GEGL, this API does not change much at all and is also the API
provided by language bindings. To make the public API available when
compiling a .c file add #include <gegl.h>, compile and link with the
flags provided by pkg-config and you should be all set. When you are
comfortable with the public API, or are using GEGL in some project
looking at the Operation reference might be useful.
The public API also includes GeglBuffer, a flexible tiled and linear
raster buffer storage system.
The bindings for use of GEGL in other programming languages than C are
co-hosted with GEGL in GNOME git but are not part of the regular GEGL
distribution. The following language bindings are currently available:
vala, C#, Python and Ruby.
The following is a small sample GEGL application that animates a zoom
on a mandelbrot fractal. compile it with:
$ gcc hello-world.c `pkg-config --libs --cflags gegl` -o hello-world
This example and others can be found in the examples subdir of the GEGL
main (gint argc,
gegl_init (&argc, &argv); /* initialize the GEGL library */
/* instantiate a graph */
GeglNode *gegl = gegl_node_new ();
This is the graph we're going to construct:
| display |
| over |
| | text |
| fractal-explorer |
/*< The image nodes representing operations we want to perform */
GeglNode *display = gegl_node_create_child (gegl, "gegl:display");
GeglNode *over = gegl_node_new_child (gegl,
GeglNode *text = gegl_node_new_child (gegl,
"color", gegl_color_new ("rgb(1.0,1.0,1.0)"),
GeglNode *mandelbrot = gegl_node_new_child (gegl,
gegl_node_link_many (mandelbrot, over, display, NULL);
gegl_node_connect_to (text, "output", over, "aux");
/* request that the save node is processed, all dependencies will
* be processed as well
gint frames = 200;
for (frame=0; frame<frames; frame++)
gdouble t = frame * 1.0/frames;
gdouble cx = -1.76;
gdouble cy = 0.0;
#define INTERPOLATE(min,max) ((max)*(t)+(min)*(1.0-t))
gdouble xmin = INTERPOLATE( cx-0.02, cx-2.5);
gdouble ymin = INTERPOLATE( cy-0.02, cy-2.5);
gdouble xmax = INTERPOLATE( cx+0.02, cx+2.5);
gdouble ymax = INTERPOLATE( cy+0.02, cy+2.5);
gegl_node_set (mandelbrot, "xmin", xmin,
g_sprintf (string, "%1.3f,%1.3f %1.3f×%1.3f",
xmin, ymin, xmax-xmin, ymax-ymin);
gegl_node_set (text, "string", string, NULL);
/* free resources used by the graph and the nodes it owns */
/* free resources globally used by GEGL */
An API to extend the functionality of GEGL with new image processing
primitive, file loaders, export formats or similar.
Each GEGL operation is defined in a .c file that gets turned into a
single shared object that is loaded. Each operation is a subclass of
one of the provided base classes:
The base operation class, which all the other base classes are
derived from, deriving from this is often quite a bit of work and
is encouraged only when your operation doesn’t fit into any of the
The filter base class sets up GeglBuffers for input and output pads
The point-filter base class is for filters where an output pixel
only depends on the color and alpha values of the corresponding
input pixel. This allows you to do the processing on linear
buffers, in the future versions of GEGL operations implemented
using the point-filter will get speed increases due to more
intelligent processing possible in the point filter class
The AreaFilter base class allows defining operations where the
output data depends on a neighbourhood with an input window that
extends beyond the output window, the information about needed
extra pixels in different directions should be set up in the
prepare callback for the operation.
Composer operations are operations that take two inputs named input
and aux and write their output to the output pad output
A baseclass for composer functions where the output pixels' values
depends only on the values of the single corresponding input and
Operations used as render sources or file loaders, the process
method receives a GeglBuffer to write it’s output into
The point-render base class is a specialized source operation,
where the render is done in small piece to lower the need to do
copies. It’s dedicated to operation which may be rendered in
pieces, like pattern generation.
An operation that consumes a GeglBuffer, used for filewriters,
display (for the sdl display node)
Base class for operations that want access to previous frames in a
video sequence, it contains API to configure the amounts of frames
to store as well as getting a GeglBuffer pointing to any of the
previously stored frames.
Used for GEGL operations that are implemented as a sub-graph, at
the moment these are defined as C files but should in the future be
possible to declare as XML instead.
To create your own operations you should start by looking for one that
does approximatly what you already need. Copy it to a new .c source
file, and replace the occurences of the filename (operation name in the
Take a look at the brightness contrast operation for a point operation
well sprinkled with comments as a starting point.
Some environment variables can be set to alter how GEGL runs, all of
these options are also available as properties on a singleton GObject
returned from gegl_config ().
A value between 0.0 and 1.0 indicating a trade-off between quality
and speed. Defaults to 1.0 (max quality).
The amount of error that babl tolerates, set it to for instance 0.1
to use some conversions that trade some quality for speed.
The number of pixels processed simulatnously.
The tile size used internally by GEGL, defaults to 128x64
The directory where temporary swap files are written, if not
specified GEGL will not swap to disk. Be aware that swapping to
disk is still experimental and GEGL is currently not removing the
per process swap files.
The size of the tile cache used by GeglBuffer specified in
set it to "all" to enable all debugging, more specific domains for
debugging information are also available.
When set babl will write a html file (/tmp/babl-stats.html)
containing a matrix of used conversions, as well as all existing
conversions and which optimized paths are followed.
Display tile/buffer leakage statistics.
Show the results of have/need rect negotiations.
Print a performance instrumentation breakdown of GEGL and it’s
operations. GEGL_USE_OPENCL: Enable use of OpenCL processing.
GEGL provides a commandline tool called gegl, for working with the XML
data model from file, stdin or the commandline. It can display the
result of processing the layer tree or save it to file.
Render a composition to a PNG file:
$ gegl composition.xml -o composition.png
Invoke gegl like a viewer for gegl compositions:
$ gegl -ui -d 5 composition.xml
Using gegl with png’s passing through stdin/stdout piping.
$ cat input.png | gegl -o - -x "<gegl>
<node class='scale' x='0.5' y='0.5'/>
<node class='png-load' path='-'/></tree></gegl>" > output.png
The latest development version is available in the gegl repository in
The following is the usage information of the gegl binary, this
documentation might not be complete.
usage: /home/pippin/src/gegl/bin/.libs/lt-gegl [options] <file | -- [op [op] ..]>
-h, --help this help information
-i, --file read xml from named file
-x, --xml use xml provided in next argument
--dot output a graphviz graph description
-o, --output output generated image to named file, type based
-p increment frame counters of various elements when
processing is done.
-X output the XML that was read in
-v, --verbose print diagnostics while running
All parameters following -- are considered ops to be chained together
into a small composition instead of using an xml file, this allows for
easy testing of filters. Be aware that the default value will be used
for all properties.
The main source of documentation as GEGL grows is the operations
reference. Plug-ins themselves register information about the
categories they belong to, what they do, and documentation of the
A link/pipe routing image flow between operations within the graph
goes from an output pad to an input pad, in graph glossary this
might also be referred to as an edge.
Directed Acyclic Graph, see graph.
A composition of nodes, the graph is a DAG.
The nodes are connected in the graph. A node has an associated
operation or can be constructed graph.
The processing primitive of GEGL, is where the actual image
processing takes place. Operations are plug-ins and provide the
actual functionality of GEGL
The part of a node that exchanges image content. The place where
image "pipes" are used to connect the various operations in the
consumes image data, might also be seen as an image parameter to
a place where data can be requested, multiple input pads can
reference the same output pad.
Properties are what controls the behavior of operations. Through
the use of GParamSpecs properties they are self documenting via
Directories marked with ⊙ use GNU make extensions to include any .c
file dropped into the directory in the build.
├──gegl core source of GEGL, library init/deinit,
│ ├──buffer contains the implementation of GeglBuffer
│ │ - sparse (tiled)
│ │ - recursivly subbuffer extendable
│ │ - clipping rectangle (defaults to bounds when making
│ │ subbuffers)
│ │ - storage in any babl supported pixel format
│ │ - read/write rectangular region as linear buffer for
│ │ any babl supported pixel format.
│ ├──graph graph storage and manipulation code.
│ ├──module The code to load plug-ins located in a colon seperated
│ │ list of paths from the environment variable GEGL_PATH
│ ├──operation The GeglOperation base class, and subclasses that act
│ │ as baseclasses for implementeting different types of
│ │ operation plug-ins.
│ ├──process The code controlling data processing.
│ └──property-types specialized classes/paramspecs for GeglOperation
├──examples ⊙ hello-world and other example uses of the GEGL API.
├──tests various tests used to maintain stability when developing
├──operations Runtime loaded plug-ins for image processing operations.
│ ├──core ⊙ Basic operations tightly coupled with GEGL.
│ ├──affine Transforming operations (rotate/scale/translate)
│ ├──generated ⊙ Operations generated from scripts (currently
│ ├──external Operations with external dependencies.
│ ├──common ⊙ Other operations.
│ └──workshop ⊙ Works in progress, (you must pass --enable-workshop
│ │ when configurig to build these by default, you can enter
│ │ the directory manually and type make && sudo make install
│ │ as well.
│ ├──external operations in the workshop with external dependencies.
│ └──generated generated operations that are in the workshop.
├──docs The GEGL website (this documentation), built for your
│ │ the corresponding sourcetree with correct documentation
│ │ etc.
│ └──gallery ⊙ A gallery of sample GEGL compositions, using the
│ │ (not yet stabilized) XML format.
│ └──data Image data used by the sample compositions.
├──bin gegl binary, for processing XML compositions to png files.
├──bindings bindings for using GEGL from other programming languages
│ not included in the tarball distribution but exist in
│ the git repository.
└──tools some small utilities to help the build.
Table of Contents
Building from source
Last updated 2012-04-02 22:57:15 BST