▲ 8 points
back
4 comments
Would using Stable Diffusion Image to Image with a low noise ratio disable most steganographic techniques? I imagine the pixels would be scattered enough to erase most embeddings. I've never had to scrub images in my career, but that's what I would try.
Yeah, there's an entire suite of comfyui custom nodes around this approach.
the article is about 'watermarking' of text documents not images
I presume all the Foundation AI companies have been doing this for years, to avoid training models on their own generated data?