/vt/ - Virtual Youtubers

>>67925132
That was Glaze that was being easily broken by LoRAs afaik. From what I read Nightshade needs a lot of images with a certain concept in them to represent another concept in order to be effective so it would make sense that it only affects big model trains.
Its working method is weird. It actually analyzes the image, tags it with 'girl' looks up in its tag database and sees that girl should be linked to 'cow' and then embeds the cow concept into the image. So all girls by all artists will have the cow concept embedded.
However it really seems they didn't keep up with some improvements as they recommend people to add the correct concept in their captions to confuse models, but I don't think any sane person would still use the image captions/alt text nowadays anyway.