Jon Sneyers

That said, it wouldn't be hard to define an optional container-level metadata box that basically contains the encoder version string and the encode options that were used. It would be...

WebP images don't have such an indication, afaik. Maybe a tool can tell you that the lossless webp codec was used, but it cannot tell you if it was used...

Many tools do mess with the image data if you save as PNG. Photoshop does use PNG8 if you check the "Smaller file" option, for example. We could show the...

> @jonsneyers > > > if it's RGB then it's likely lossless > > Can current version of reference implementation create lossy RGB via available API, or it's just theoretical...

Yes, same as with *.png, basically.

> > Yes, same as with *.png, basically. > > In internal files of mobile software and videogames? Sure, and 8-bit palette speaks by itself anyway. In art sites? Photography...

Yes, that makes sense. But it _is_ metadata, not an inherent property of the codestream. In libjxl, we make a sharp distinction between image data and metadata. Things like ICC...

Do you mean the `pHYs` physical image dimensions that PNG files can have? We don't have such a thing in the JPEG XL codestream headers; we did not consider it...

In this case, the problem is too many extra zero runs: spec limit is 65555, while there are 149279 of them in this jpeg. Again, as in the case of...

The spec cannot be changed, but it might still make sense to have an encode option to basically do the `jpegtran` thing for you, i.e. produce a `jbrd` that doesn't...