Are markers being removed here the same or similar to ones tools might add if you use an AI tool just to edit a photo? like a more complicated object removal in a photo editor?
I feel like it is even worse if we have a marker that is ALMOST always present on AI images. It would make us more likely to be fooled by an AI image that had them removed, because we would trust the marker.
This is a bit misleading as for Gemini it only properly removes the visible watermark. To remove SynthID it has to regenerate the image at low noise with SDXL, which will likely destroy a lot of small details, plus won't work for higher res properly (NB2 and GPT Image 2 support up to 4K image outputs)
Nano Banana 2 only supports 1K resolution (1024x1024) natively. Anything above that is upscaling. So this matches SDXL. GPT Image 2 does support 4k natively (but experimentally).
> Use cases where the threat model fits: You are preserving art or historical record against false-positive "AI-generated" labels.
Sorry, how does using AI to generate images have anything to do with this? Image generators cannot insert watermarks into things they did not generate, and it seems highly unlikely that you will get a false-positive watermark on human-generated art, especially if, as the readme says, these watermarks have high enough fidelity to trace to a specific session id. Plus the modifications to the image needed to erase watermarks would necessarily change the thing being "preserved."
[edit]: the more I read the more I'm convinced, the claimed use cases in the README are bullshit and the real reason is to provide a tool that helps people bypass "AI-generated" labels on social media for AI slop.
There's an underappreciated comment in the other thread about SynthID and OpenAI [0] that captures what (IMO) the hacker ethos on this should be. We care about privacy, we should not accept tools that barcode our every digital move. (note that the counter of "well, they don't do that yet" is not particularly convincing)
I just saw the announcement about OpenAI or so going to use SynthID and all I thought was; what can d be read(located) can be removed. Seems the tool already exists, proving my point.
Regardless of one's opinion about this particular project, it seems obvious to me that the path forward is proving authenticity of non-AI resources rather than attempting to watermark all the AI-generated ones.
watermarking only really works when the scheme is secret.
putting cyphertext in high frequency noise is old news. in generative land would be far more interesting to use the generative flexibility to encode in macrostructure.
Watermarking images generated from trained data on stolen copyrighted material, I get why so they can try to tell if something is real or not but something seems wrong
Why? Don’t do this. Society is built on an implicit assumption of trust. You will erode the foundation that enables any success you might have in the short term.
I think AI watermarks are kind of a lost cause anyway.
I'd be more interested in some kind of trusted Non-AI watermark.
This is something that could get integrated into cameras for example. However, considering how much AI-processing we already have in "normal" photos, it will be difficult to decide where to draw the line.
44 comments
[ 2.4 ms ] story [ 56.1 ms ] threadSorry, how does using AI to generate images have anything to do with this? Image generators cannot insert watermarks into things they did not generate, and it seems highly unlikely that you will get a false-positive watermark on human-generated art, especially if, as the readme says, these watermarks have high enough fidelity to trace to a specific session id. Plus the modifications to the image needed to erase watermarks would necessarily change the thing being "preserved."
[edit]: the more I read the more I'm convinced, the claimed use cases in the README are bullshit and the real reason is to provide a tool that helps people bypass "AI-generated" labels on social media for AI slop.
[0]: https://news.ycombinator.com/item?id=48200060
- Rocky
putting cyphertext in high frequency noise is old news. in generative land would be far more interesting to use the generative flexibility to encode in macrostructure.
I'd be more interested in some kind of trusted Non-AI watermark.
This is something that could get integrated into cameras for example. However, considering how much AI-processing we already have in "normal" photos, it will be difficult to decide where to draw the line.