Mairo said:
One of sites goals is to archive, so of course we want to have original Mona Lisa, instead of photograph of Mona Lisa taken with mobile phone and printed on dot matrix printer. We want to make sure that images shared today will be as good as they are in next 10 years, instead of someone only being able to find compressed version from iFunny.
Many websites do also resave the content, so if you save JPG as JPG, now you are doubling the compression, where PNG as PNG keeps visual quality identical.
Also because PNG files are still usually couple megabytes max, modern internet connections should handle them with ease, so benefit of lower filesize isn't enough to justify altered and worse quality. Also like Chessax said, there are even software dedicated for further optimizing these files without any quality loss, so if you can save your artwork as PNG, it's usually best approach.
Twitter, tumblr, facebook, etc. nature is to share something right now, were it memes or selfies, memes don't matter if they are bit compressed and real life photograps it's harder to see JPG compression, so for social medias it's far more important to make content load fast to users and preserve bandwidth and server space over long time archival.
We care about these things, that majority won't have to.
Reason why samples are enabled for guest viewers (on top of lessening server load) and samples and thumbnails are JPG.
And this actually comes to good point, JPG is still good if you only want to see the image, but as that can and is generated from original content, original content should still be as good as it possibly can be. Those with low end phones and connections get the content, those wanting original quality get the content, win/win.
Ah, yeah, as usual what I think I say isn't always what I come across as. You're right about PNGs being lossless and all that and perfect for archiving, I was trying to be more general as in using PNG24 (or even PNG32) for thumbnails and even samples/previews, e.g. in automated thumbnailing systems (like on e621), can be kind of over-kill, i.e. both formats have merit albeit in different places and situations.
Since sites seem to like hiding their high quality versions, as you mentioned, I could see why highly detrimental resampling happens more often than not.
I would go as far as to say that the biggest problem hasn't even got to do with quality but resizing, e.g. resampling a JPEG with the same moderately high settings (90) iteratively 100(!) times will show very little deterioration in quality (to the point of sometimes not being discernible), and even doing the same for a relatively low quality setting (50) will not give anywhere near a 100 time decrease in quality, it's basically the first (or lowest) resampling that determines quality.
However, decreasing and increasing size by just a bit will have major consequences to sharpness (unless you're very smart about it). Something our precious PNG format still suffers from. So the problem is just as much about resizing, possibly in combination with using low quality JPEG which will exacerbate the problem further.
Also tried to point out that if one uses low or (even moderate to high) quality JPEGs for full images then one is either an idiot or ignorant, note that being ignorant is perfectly acceptable, albeit not ideal. One should either use a lossless format (which you advocate) or crank the quality pretty much all the way up too 100 (which I find acceptable as well).
Also got samples turned on on e621; yay for fast page loads and less memory usage (don't really care about bandwidth as I got more than plenty).
Mairo said:
PNG ensures that it's visually original, where even highest quality JPG is already altered.
e621:image_quality
https://xkcd.com/1683/
What is up with that 100% JPEG? That's the worst JPEG resample at that quality that I've ever seen in my entire life. That can't be a proper 100% JPEG (maybe the software is using a poor quantization matrix?), because if I resample that PNG to JPEG using ImageMagick I can't tell the PNG and JPEG apart even when zoomed in to 16x, but the example image looks horrendous in comparison. I tried in Photoshop too and it did pretty much just as well. Maybe I need to reevaluate just how bad some tools are...
Seems we're getting a bit off topic again, nothing unusual when I'm around...
Updated by anonymous