Alright, I have been summoned here to shed some light, so let me clarify some things.
When opening SG's original images (5K res) through a hex editor and looking at their header, you could see that they weren't coming from ClipPaintStudio or Photoshop like the old ones, but they had some weird data before the JPG marker for "Frame Start.".
Moreover, if you ran them through a good stego scanner they would come up as PixelKnot(guardianproject/PixelKnot) signature, which is not necessarily accurate, but it meant SG was using F5 steganography.
F5 stego is something you can get rid of by just re-compressing the image on a lossy encoder like .JPG. That will scramble all the numbers and ruin the encoded bits.
I'm pretty sure this was SG's first attempt to add tracking, and it didn't really work largely because most web services re-encode images to save bandwidth, so they added one more layer: hidden watermarks in the images.
^ This was what got the majority of the people banned at the time, and was why I stepped in to help.
The watermarks were initially very small and used a contrast blending mode so the outlines of the text were lighter than the inside.
That was also easy to remove, which is why I released a script that could do that automatically, so there was no need to waste time and delete the codes 1 by 1 using magnifying glasses, microscopes, paint brushes or whatever the
gatts2 is using nowadays.
The method was documented in the python files and worked by dissecting the images into individual bitplanes and deleted the tracking codes by changing the LSB(least significant bits), which is the data F5 stego tools change in order to add short hidden messages with minimal image quality loss.
In addition, it also detected the actual watermarks and automatically masked those areas with AI content aware-fill to regenerate some sectors of the images, then added more noise to make the watermarks unreadable in case the script missed one (selective blur, grain, scaling, etc.).
This method worked flawlessly and was used for months by me, Tron and others without any issue whatsoever.
SG was well aware of the situation and was making changes to the way the images were tracked every few months, for instance when they figured out that part of the process was to do a 50% downscale, they doubled the size of the watermarks, so I had to make changes too.
At some point I stopped helping with the leaks on the old forum because one user got banned in, seemingly, an automated way. Specifically, they got banned 1h after posting some images that were cleaned using my old method, and although they weren't cleaned by me so I can't vouch for them to be cleaned properly, the ban was too strange and I couldn't figure out how it happened based on the information I had available at the time.
One possible theory, was that SG or his minions were leaking user IP's from the forums so they could match them with their subscribers downloads from ExpressCDN. I personally think that was too elaborated and according to Occam's razor, a simpler explanation is more likely. In this case, it's more likely that there's a more straightforward reason for the bans without involving multiple steps and assumptions. In the end I didn't bother looking into it because there weren't any more breadcrumbs to follow.
Hopefully this wasn't too technical and will be helpful to anyone who wants to attempt solving SG's puzzles in the future.
o7