AI disturbance overlays for those who don't have Ibis paint premium. found them on tiktok
how do you use these?
Put these on the top layer above everything, set layer to 'overlay' then adjust opacity. You can put it on whatever opacity you want but usually 30%+ is most effective.
The point is to obstruct the picture so AI can't read your image because AI counts every single pixel in your art
By this post alone, in less than 5 hours.
I need you all to calm the fuck down
NEEDED AND NO WE WONT THIS IS BRILLIANT
Hate to be the one to break it to y'all but uh... This is more tiktok bunk.
It is absolutely not going to prevent a model from being trained effectively on an input image by setting a semi-opaque noise pattern over it. The entire thought process is flawed??? Deeply. Models don't exactly "count every pixel"???? [ where is this information coming from??? This is insane. ] and even if they did, shifting each ones color by a semi-arbitrary amount, leaving the art itself visible, is not going to do anything but cause any renders based on labels used for your style [which is probably not even being bothered with at this point, unless you're incredibly talented or draw really novel porn] to possibly be shidded out with a similar noise pattern. (though, honestly, training a model to output in a way that ignores these patterns would be trivial, especially since they've been posted publicly and consist of a very small sample size. Literally imagemagick and thirty minutes of scripting with a dataset of images to duplicate with these overlays would be all that's required to generate a dataset to counter even the tiny amount of noise these are introducing. And that's to entirely ignore the fact that training a LORA on custom art is so trivial anyone could do it. )
Also, the "Ai" is not sniffing the entire internet, eating every image automatically, and training against it. Datasets have to be collected, curated, labeled, and built, in a multi-step process, even today. That's why uploading "poisoned" art doesn't work.
Your understanding of the technology you are trying to wield amounts to a clumsily shoveled together superstition based on here-say and misunderstandings of diffusion as a process.
I did my own breakdown on this, on this very thread, but the reply above gets to the point very well.
Clumsily shoveled together superstition, indeed.
