That is only applicable to third party sources and non-sourced posts. The sources here are all from the artist and are original.
Source, in this case, in the sense of where the image was posted, not who posted it. Twitter's compression algorithm is harsher on image data than Pixiv's, so Pixiv images come out larger, with less artifacting and data loss.
For the sake of clarity between platforms, I'll quote my reply on Discord:
It isn't about what's visible, though. It's more like what's closest to the most recent version the artist exported from their drawing program. Revisions are highest in the chain because they're "more recent," then for images that were obviously identical before being fed into the internet data rape machine, that which is least raped takes precedence.
The images are essentially visibly identical, but Twitter's an asshole and changed the image along all of the edges. This wasn't done by the artist, so it becomes the child post.
Both versions have edge artifacts plainly visible when looked at 400% magnification. I don't view my images at 400%, but even then, there is not much difference, so take your pick.
Now run either image through waif2x at Artwork, No upscaling and Low noise reduction settings and those artifacts vanish with no visible modification to the image colors even at 400% saved as a 500k png.
I won't say it will clean up even the poorest images, but it really does a fine job.
I wish I could upload that, but I don't know if that runs afoul of Danbooru's rules about altering artwork. But if the artifacts are visibly annoying, that's a good way to clean it up.
