[bflesch]

Can you explain a bit how working with JPG format was different than working on PNG format? How did you achieve significant savings for both formats, given that many clever people have worked on this before?

[iambavith]

We focused on lossless techniques like metadata cleanup, optimal encoding settings, and structural compression—for both JPG and PNG.

[CyberDildonics]

Does this mean that some of your 'compression' is just deleting meta data?

structural compression

What does this mean?

[iambavith]

Yes, some gains come from stripping metadata, but structural compression means optimizing how image data is encoded, like better Huffman tables for JPGs or smarter filtering for PNGs.

[CyberDildonics]

some gains come from stripping metadata,

That's not compression, that's just deleting people's data.

structural compression means optimizing how image data is encoded, like better Huffman tables for JPGs or smarter filtering for PNGs.

I think that's just called compression.

How are you getting better huffman tables for JPGs and how are you doing 'smarter' filtering for PNGs ?

I'm asking for deeper technical explanations because not only would beating current image compression libraries be a technical feat, a lot of people in this thread think that you aren't actually doing anything differently.