Convert any image into a low-resolution, color approximated textual representation for use on Discord.
Discolor is a converter for images that scales them down and converts them into
strings of spaces and full blocks with control sequence graphics rendition
(ANSI)
codes to change the text color. The text images can be displayed in Discord code
blocks by using ansi as the language (currently only supported on
browser/desktop).
usage: discolor.py [-h] [--width WIDTH] [--height HEIGHT]
[--char-ratio CHAR_RATIO]
[--quantizer {pil_floyd,pil,distance,direct}]
[--scaler {nearest,box,bilinear,hamming,bicubic,lanczos}]
image
Convert images into Discord-compatible text art.
positional arguments:
image
optional arguments:
-h, --help show this help message and exit
--width WIDTH Target width, if not specified, will be determined by
height.
--height HEIGHT Target height, if not specified, will be determined by
width.
--char-ratio CHAR_RATIO
Character ratio, corrects for characters being taller
than they are wide.
--quantizer {pil_floyd,pil,distance,direct}
Algorithm to convert colors to the palette. Use direct
for images already in the palette.
--scaler {nearest,box,bilinear,hamming,bicubic,lanczos}
Downscaling algorithm. Use nearest for pixel art.
Convert pixel art made with the palette:
./discolor.py --quantizer direct --scaler nearest sprite.png
Convert a photo:
./discolor.py --width 30 photo.jpeg
Convert a photo and copy it to the clipboard:
./discolor.py --width 30 photo.jpeg | xclip -selection clipboard
Discolor first scales the image to an appropriate size. The size is determined by user input, as well as (if one of width or height is not specified) the aspect ratio of the characters used. This ensures pixels are square, even though the underlying characters used are rectangular.
Once rescaled, it is quantized, reducing the color depth to match the amount of colors available in the palette. This step may involve dithering, which adds colored noise to blend colors into one another to attempt to increase the perceived color depth.
Finally, the converted textual pixels are catenated together. This process is smart, and makes sure that only changes to the color of the textual pixel produce a corresponding escape code. Further, only escape code transitions that require a reset (0m) will get one. Together, these greatly reduce the amount of characters needed to display a given image.
Ideally, your image will have as little transitions as possible. A transition is when one color is changed for another along a row of pixels. For instance, if a row alternates between blue and red every other pixel, it will have a lot of transitions, whereas a solid green row will have none.
The catenation process can reduce the amount of characters used for a given row, but only if it has few transitions. Rows of the same color can use the same color introducer. Making your image use long stretches of the same color will produce a smaller output than if it had transitions.
If you have an image that is not pre-quantized/posterized, the pil and
distance quantizers will produce output with the fewest transitions. Since
pil_floyd uses dithering, it will produce many transitions. For images with a
lot of color, this dithering process could greatly increase the final size.
For the best results, quantize your image before conversion to text. This can be done by using the GIMP color palette and running a quantize or posterize with that palette. You can experiment until you get a result with a good balance of quality, size, and transitions. This can then be converted to text with:
./discolor.py --quantizer direct --scaler nearest image.png
The direct quantizer directly converts colors into the palette, making invalid ones pink/magenta. The nearest scaler prevents new colors from being added to the final result.
Public domain (Unlicense).