umbra VS zopfli

Before I went on my Christmas vacation last year I wrote an article on how I use Nix in my Elm projects. At the time, I was pleased with my set up. However, not even a month would go by before my satisfaction was questioned. In early January, Carlo Ascani asked a question, on the Elm Discourse, about his Umbra project. I decided to explore his project and I soon discovered two files, devbox.json and devbox.lock, I had never seen before. This piqued my curiosity and I had to learn more. I followed the link to the Devbox website and feverishly read the docs. I... was... hooked. I was pleasantly surprised by its simplicity and it seemed to fit my use cases really well.

This project uses HTMLMinifier, optipng, and zopfli to create a custom production Elm build pipeline. You can see how I make use of these tools in this build script. Here are the results in case you're interested. I used the same ideas from this project to build and deploy dwayne/elm-conduit, which you can learn more about in my article Yet Another Tour of an Open-Source Elm SPA.

I'm making a 2D game with lots of PNG assets. I recently found that ZopfliPNG regularly reduces the sizes of my assets by ~3x compared to Asesprite's output. I've tried a few other PNG compression tools, but ZopfliPNG consistently outperforms the others.

I mean something better than Google's best engineers trying to optimize LZ77's compression as much as humanly possible, while remaining compatible with the DEFLATE/zlib bitstream.

zopflipng typically beats pngcrus and optipng (on Linux at least) but by default it drops auxillary PNG chunks [0] which can result in browsers (and other applications) using a different color space, causing the resulting images to look more washed out than the original. To prevent this you need to explicitly pass --keepchunks=cHRM,gAMA,pHYs,iCCP,sRGB,oFFs,sTER to zopflipng.

Unfortunately it (and most other tools) don't have APNG support, keeping only the first frame.

[0] https://github.com/google/zopfli/issues/113

I tried going with a lot of different compression algorithms, in the end a lot of iterations of Zopfli delivers the best result. With that, I got the image down to 591 kB.

I tried zopflipng [1] with the very expensive option suggested from the usage and got 2,493 bytes (original 2,740 bytes, nowhere practical as it took 3 minutes) so you don't have to sacrifice the palette, though I'm not sure if 16 or even 4 color palette will significantly alter the visual.

[1] https://github.com/google/zopfli