This is a tool for making images have a particular palette of colors. Simply give image-factory a palette.txt file full of the desired colors, and it will give you an image containing only those colors.
| Original (3648x5472) | Nord Color Pallete (2.688 seconds) |
|---|---|
 |
 |
There are three mandatory options, one for an input image, one for a palette.txt file and one for an output destination.
image-factory --input wallpaper.jpg --palette palette.txt --output new_wallpaper.jpgOther flags can be seen with the built in help image-factory --help.
View the palette.txt file in this repo for an example of the Nord color palette.
I'm currently working on packaging it for Nix. I'm happy to help with packaging for other package managers, but only have the knowledge for Nix.
For the time being though, Nix users (with flakes) can run image-factory like so:
nix run github:IllustratedMan-code/Image-Factory -- --helpopencv4pkg-configcli11bear(optional)ccls(optional)
After cloning (or forking) this repo, run make. This should create an executable called image-factory.
Nix users can just run nix develop to get the required dependencies, nix build to build, and nix run to run.
How does image-factory work? I'm using the simplest euclidean technique outlined here.
The closest color in the palette is calculated for each pixel in the target image using the relative distance formula in three dimensions.
The relative weight of each color is defined by scalars r, g, and b.
This is the naive (least efficient and least complex) approach to the nearest neighbor search problem in a coordinate system.
This shouldn't be a problem for most use cases, but as palettes grow in size, image-factory will get slower in a linear fashion.`
All of these improvements will add complexity to the program.
A space partitioned tree structure might help here, but it won't neccesarily perform better.
This method uses a graph data structure to determine distance. It is not 100% accurate, but it may be faster than the current implementation
A three dimensional Delauny triangulation or more aptly a Delauny tetrahedrization could be used to determine the closest four colors in the palette.
An Opengl shader would probably outperform all of these algorithms because it takes advantage of the gpu for calculations. Its also a lot harder to implement (at least for me).
This is a rust shader library, see example here.
It appears that opencv can be compiled wit CUDA support which could make image-factory faster
ImageGoNord is the most popular alternative, gruvbox-factory used ImageGoNord behind the scenes.
It's a much more complicated program with web APIs and other such niceties. image-factory is much simpler, but I believe the underlying algorithm is similar as output looks very similar.
I believe image-factory is faster, but I haven't done detailed testing.