The simplest way to beat any stenography protection would be to downsize the image by %50, then use a readily available AI upscaler like chainner with a model like 4k_booruGan, then downsize by %50 again to end up in the original size, any embedded data in pixels will be outright fcked by all the downsampling and AI upscaling, with minimal quality loss to the art itself, this is what was done on that Ai Papi #20 upload.
If the text gets fked (like in Ai Papi 20) then there's upscaling models made specifically for text, would be a matter of doing a separate upscale for that and then merging them, given he makes like 5 pages a month it's not a massive amount of work.
If the text gets fked (like in Ai Papi 20) then there's upscaling models made specifically for text, would be a matter of doing a separate upscale for that and then merging them, given he makes like 5 pages a month it's not a massive amount of work.