Hacker News new | ask | show | jobs
Google's SynthID watermark is hard to break, but doesn't solve AI disinformation (arstechnica.com)
3 points by thm 2 days ago
1 comments

If you are experienced working with diffusion models it's rather easy to break. A simple img2img pass with low denoise will generally do the trick. It's like locks for normal people they are a good enough deterrent but if someone know what they are doing they are easy to break.