2y ago

Google debuts a new way to watermark AI-generated images

www.axios.com

Just a moment...

Google's DeepMind unit is unveiling today a new method it says can invisibly and permanently label images that have been generated by artificial intelligence.

14 comments

- You heard of stable difusion? They got 1 line installs nowadays then all you have to enter is a prompt and go.
  Entirely open source so anyone could improve the model or not, and it'd be more than legal to release a non watermarked version (if a watermarked version even ever appeared).
  I saw down the chain it was compared to deuvono, which I'd argue is a bad analogy - cause whos gonna run a rootkit on their PC just to create an image, especially when there's a million options not to (unlike games which are generally unique)
  
  This is Google were talking about. We're probably going to find out that you can remove the mark by resizing or reducing the color depth or something stupid like that. Remember how YouTube added ContentID and it would flag innocent users while giving actual pirates a pass? As said in a related article:
  "There are few or no watermarks that have proven robust over time," said Ben Zhao, professor at the University of Chicago studying AI authentication. "An attacker seeking to promote deepfake imagery as real, or discredit a real photo as fake, will have a lot to gain, and will not stop at cropping, or lossy compression or changing colors."
  https://www.maginative.com/article/google-deepmind-launches-new-tool-to-label-ai-generated-images/
TBF, I don't think the purpose of this watermark is to prevent bad people for passing AI as real. It would be a welcome side-effect but that's not why google wants this. Ultimately this is supposed to prevent AI training data from being contaminated with other AI generated content. You could imagine if the data set for training contains a million images generated with previous models having mangled fingers and crooked eyes, it would be hard to train a good AI out of that. Garbage in, garbage out.
- So theoretically, those of us who put original images online could add this invisible watermark to make AI models leave our stuff out of their "steal this" pile?
  
  Yea actually, that has a good "taste your own medecine" vibe
- AI-generated images are becoming increasingly realistic, AI can't tell them apart anymore.
  
  iirc AI models becoming worse after being trained with AI generated data is an actual issue right now. Even if we (or the AI) can't distinguish them from real images there are subtle differences that can be compounded into quite large differences if the AI is fed its own work over several generations and lead to a degraded output.
- I’m not sure that’s the case. For instance, a lot of smaller local models leverage GPT4 to generate synthetic training data, which drastically improves the model’s output quality. The issue comes in when there is no QC on the model’s output. The same applies to Stable Diffusion.
says so in article:
The watermark is part of a larger effort by Google and other tech giants to develop ways to verify the authenticity of AI-generated images.
This is important because AI-generated images are becoming increasingly realistic, and it can be difficult to tell them apart from real images.
Spoiler - they will secretly have all humans in ai generated art have slightly messed up hands.
Mind blown!
Its invisibility should really help all the laypeople see it clearly.

14 comments