itch.io

Great! 

Content-Adaptive Image Downscaling is especially interesting, but Pixelated Image Abstraction (Gerstner et al. 2012) seems to give good results on photos (especially if one wants a pixelated result).

There are some code examples on github if that would be helpful:

-Content-Adaptive Image Downscaling (Matlab):
https://github.com/AyushRai09/Content-Adaptive-Image-Downsampling
and pseudocode:
https://johanneskopf.de/publications/downscaling/paper/pseudocode.pdf

-Pixelated Image Abstraction (Gerstner et al. 2012) (Python):
https://github.com/gskaggs/Pyxeled

Just be aware that the Content-Adaptive Image Downscaling algorithm is not designed for palette reduced results on natural images:
https://johanneskopf.de/publications/downscaling/supplementary/html/gerstner_com...

I second this; focus on turning art into pixel art. I have tried 5-6 similar programs to yours and they all struggle with producing great results, especially when downscaling photos and/or downscaling to small sized sprites (64x64 pixels or less) which often end up as a blurry or pixelated mess. Vector graphics gives reasonable good results. It's not without reason that image downscaling is one of the most classical problems in computer vision ...

I would therefore suggest adding several different downscaling algorithms (preferably as many as possible):
https://en.wikipedia.org/wiki/Image_scaling#Algorithms

You might want to take a closer look at the following methods as they seem to provide great results when downscaling to small sized pixel art:
1. Content-Adaptive Image Downscaling:
https://johanneskopf.de/publications/downscaling/
2. Pixelated Image Abstraction (Gerstner et al. 2012):
https://gfx.cs.princeton.edu/pubs/Gerstner_2012_PIA/index.php
3. (and maybe) Subsampling and Bicubic sharpened.

Here is a comparison of all 3 methods:
https://johanneskopf.de/publications/downscaling/supplementary/pixel_art/index.h...
https://johanneskopf.de/publications/downscaling/supplementary/html/gerstner_com...

Having adjustable parameters for some of the algorithms would also be nice so that the downscaling can be fine-tuned.

Another interesting idea would be to add a cel shader to the program. This can be used to "flatten" the image (especially on photos) or smooth out blurriness to get a more vector-like image which again might produce better downscaling results.