Release of G'MIC 4.0

David_Tschumperle · February 16, 2026, 10:39am

This is the changelog for the release of the 4.0 version of the G’MIC software, released on 2026/07/01.

It lists all new features and changes done since version 3.7.

What is G’MIC?

G’MIC (GREYC’s Magic for Image Computing) is a full-featured open-source framework for image processing. It provides several different user interfaces to convert/manipulate/filter/visualize generic image datasets, ranging from 1D scalar signals to 3D+t sequences of multi-spectral volumetric images, thus including 2D color images.

G’MIC main project page : https://gmic.eu
G’MIC download : G'MIC - GREYC's Magic for Image Computing: A Full-Featured Open-Source Framework for Image Processing - Download
G’MIC web service : https://gmicol.greyc.fr/
G’MIC language tutorial pages : Start Here

^ The plug-in G’MIC-Qt in action, running on GIMP 2.10 ^

What’s new in version 4.0?

New features

G’MIC-Qt plug-in:

[gmic-qt-400] New filter Rendering / Pixel Stretch adds colorful streaks to your images along user-defined spline curves.

[gmic-qt-400] New filter Colors / Transfer Color [Multi] proposes several different methods to transfer colors from a style image onto a content image, possibly with user guidance.

[gmic-qt-371] New filter Artistic / Marker Drawing applies a marker drawing effect on selected images.

G’MIC Standard Library:

[stdlib-375] New command output_csv saves arrays of strings or numbers as a .csv file.
[stdlib-371] New command unitnorm computes the pointwise Lp-norm in selected images.

Others:

[web-373] New code repository gmic-interactive-demos provides sources of some of our interactive demo programs (used at booths for Science Day)

[web-371] Added a new code sample “Rotating Diffusion-Limited Aggregation”, in the Gallery section of the G’MIC website.

[emacs-371] Created a gmic-mode.el file, that defines a G’MIC mode for Emacs : Automatic indentation and syntax coloring for G’MIC scripts.

Improvements / Modifications

G’MIC Standard Library:

[stdlib-400] Command rbf : Additional affine term can be added in the interpolating function.
[stdlib-400] Command transfer_rgb has been recoded from scratch. It implements a better variational algorithm and is way faster to run.
[stdlib-375] Command split_colors has been recoded from scratch and is faster.
[stdlib-375] Command input_csv has been recoded from scratch.

Core functionalities:

[core-400] Command qr: New parameter to use pivoting in QR decomposition, and return the permutation vectors of matrix columns.
[core-400] New mechanism to auto-check validity of a shared image when it is used in a selection or as an argument. Slightly slows down the code but avoid nasty code that can segfault.
[core-400] Syntax {``string} has been replaced by {:string}.
[core-400] Command displacement has been recoded from scratch. It is now faster, more precise and better manage prescribed vector constraints.
[math-core-376] Optimization of variable assignements, so that an expression like X = cos(X) is actually computed “in-place”.
[math-core-375] Optimization of functions when working on vector-valued arguments (e.g. cos(X) where X = vectorN()).
[math-core-375] New functions ‘_cumulate()’, ‘_equalize()’, ‘_mirror()’, ‘_noise()’, ‘_normalize()’, ‘_permute()’, ‘_reverse()’, ‘_shift()’ and ‘_sort()’ define in-place versions of their equivalent functions (whose names do not begin with an underscore).
[math-core-371] New pre-defined variables _0 and _1 which are vectors whose dimension is equal to the number of channels of the associated image, and whose values are all 0 or 1.
[math-core-371] Optimization of vector initializers, with smaller generated opcodes.

Others:

[web-373] Successfully upgraded the G’MIC-Qt plug-in for GIMP 3.2

Bug fix:

[stdlib-375] Fix bug in command remove_duplicates.
[core-372] Fix UB in G’MIC interpreter, caused by trying to unlock a mutex that has not been locked before.
[core-371] Fix line drawing algorithm bug happening when specified line coordinates are very large (out-of-image), but a when portion of the line should be drawn anyway in the image.
[math-core-400] Fix regression introduced in 3.7.6: Check correctness of lvalue when assigning variables.
[math-core-400] Fixed expression involving variable assignment and pixel modification, like var = i(x,y) = value (was not assigning image pixel).
[math-core-371] Variable $% were returning NaN in a single-iteration loop, when accessed through the math parser. Now returns 0 .
[math-core-371] Variable $var were returning 0 instead of NaN if $var was as a string rather than a number.
[math-core-371] Fix bug in SVD computation that was computing wrong singular values in some cases. That was affecting the functions svd() and eig().

Reptorian · February 16, 2026, 1:19pm

What would I use null for?

David_Tschumperle · February 16, 2026, 2:02pm

A typical use is to tell eval to perform the image loop only over axes x,y,z, and not all axes x,y,z,c.

Example: The following code will produce a binary image that has values 0 or 3.

$ gmic sp colorful 100%,100% eval.. "max(I>128)?++i(#-1,x,y,0,0)"

Whereas, this one:

$ gmic sp colorful 100%,100% eval.. "max(I>128)?++i(#-1,x,y,0,0); null"

will produce an image that is either 0or 1.

(It’s equivalent to : $ gmic sp colorful 100%,100% eval.. "max(I>128)?++i(#-1,x,y,0,0); I" , but null costs nothing, while I takes one operation per pixel).

Reptorian · February 16, 2026, 2:10pm

I see, so, in that case, wouldn’t it makes sense to replace I with null? I think I’ll do that in my code when I get to doing it.

David_Tschumperle · February 16, 2026, 2:13pm

Yes, it would make sense.
What I do sometimes, to avoid putting I; at the end of the fill or eval expression, is add a begin(ret = vectors()) section at the beginning of the expression, then use ret as the very last item of the expression.
But null will simplify this.

Reptorian · February 16, 2026, 7:03pm

Got it. This does raise a question though. I haven’t tested it yet, but this should work fine?

if(const_variable
,do_something(); # const_variable is true, then do something
,null;           # Otherwise null
);

EDIT: Looks like I can’t test it unless I compile. So, when’s the prerelease repository is going to be updated?

David_Tschumperle · February 17, 2026, 7:10pm

It’s already there, in branch develop:

Reptorian · February 17, 2026, 7:19pm

I mean here: Index of /files/prerelease

lylejk · February 17, 2026, 10:11pm

Now that’s just too cool, David.

David_Tschumperle · February 21, 2026, 5:08pm

@Reptorian , just FYI : I’ve built pre-release binary packages (3.7.1_pre) and uploaded them on the G’MIC server.

David_Tschumperle · February 25, 2026, 7:34am

@Reptorian , finally there are some changes concerning null.
It is replaced by _0 , which is a vector of 0 , having the size of the spectrum of the associated image. I’ve also added _1, that is the same but with 1s instead of 0s.

Reptorian · February 25, 2026, 7:41am

Doesn’t this affect things like const _0=_'0';?

I use that construct for in case I need int representation of char, and underscore tells me it.

David_Tschumperle · February 25, 2026, 7:43am

It’s like any other predefined variables, you can ovveride it with your own definition if you want, so const _0 = _'0'; is still valid and working.

David_Tschumperle · February 26, 2026, 1:02pm

2026/02/26: Release of G’MIC 3.7.1.

Now taking a short break to recharge, hopefully to be back more motivated!
Enjoy!

lylejk · February 26, 2026, 3:28pm

Haven’t tested much, but had to try the Marker preset, David and it works great. With a bit of Convolution stuff to bring some of the details back, the below is a result. Much improved over my actual capture. lol

David_Tschumperle · February 26, 2026, 3:31pm

Great job, Lyle!
Honestly, this filter is really becoming my favorite at the moment. It ultimately has relatively few parameters to adjust compared to the wide range of results it can produce.
From an algorithmic point of view, I find that truly remarkable

lylejk · February 26, 2026, 3:33pm

re"marker"able for sure, David. lol

David_Tschumperle · February 26, 2026, 10:29pm

2026/02/26: Release of G’MIC 3.7.2.

Yes, only a few hours after 3.7.1 was released, but I spotted an undefined behavior in the interpreter code, and I really wanted to fix this before any problems arose!
Really sorry about that.

Reptorian · March 4, 2026, 11:55am

I can’t keep this question out of my head, but do you think 0x$copies can be done in parallel? I think this would be a good idea to improve execution time code that use dictionary structure. Like imagine it can be done in parallel, and instead of low values, you can create 100000+ 0x0x0x0 without it being a bottle neck.