There was a pretty popular video showing this exact concept (hiding shapes/text in high frequency noise ) about a year ago, for those interested:

https://www.youtube.com/watch?v=Bg3RAI8uyVw

Can AI vision models process "Magic Eye" images? the demo video reminds me of text in SIRDS images.

A different link: https://www.mixfont.com/ghost-font

Please note - I’m the submitter, not the creator

Earlier: https://news.ycombinator.com/item?id=48870381