>>106464001
glaze/nightshade doesn't really work
here are the conditions to maybe see an effect in training
>have to use the exact same model that the adversary noise was trained for
>have to make no transformation to the image before training (resizing, blurring, antialisasing, de-edge or de-jpg...)
>have to use the trained lora on a very similar model
and the most important
>have to crank up the adversarial noise a LOT which will make very noticeable artefacts and reduce the quality of the work
but I do agree that all online services should shut down, as the main producers of garbage slop (and slop patreons) are subcontinentals too poor for gpus