On the “better compression” front, I’d also add that I doubt that either PNG or WebP represent the pinnacle of image compression. IIRC from some years back, the best known general-purpose lossless compressors are neural-net based, and not fast.

kagis

https://fahaihi.github.io/NNLCB/

These guys apparently ran a number of tests. They had a neural-net-based compressor named “NNCP” get their best compression ratio, beating out the also-neural-net-based PAC, which was the compressor I think I recall.

The compression time for either was far longer than for traditional non-neural-net compressors like LZMA, with NNCP taking about 12 times as long as PAC and PAC taking about 127 times as long as LZMA.