Posted by smooke 8 hours ago
I have found great success of getting rid of it by masking every 2nd pixel, regenerating missing pixels and then once again masking every 2nd pixel offset by 1.
Used an off the shelf model to fill in the pixels, but I also exported a depthmap first (before any alternations) and denoised it so generated masked pixels comform to the original content. The result was obviously not 100% perfect, but with more time and a model fine tuned for this specific use-case would be able to remove any kind of ai watermarking without too many issues.
Always amusing to see AI used against itself.
https://deepwalker.xyz/blog/bypassing-synthid-in-gemini-phot...
Can it be used to create something like nutritional labels for synthetic content? 10% synthetic text, 30 synthetic images.
Your reality was 15% synthetic today (75% mega corp, 25% open-weight neocloud).
Presumably the deployed version is meaningfully different.
https://github.com/swesterfeld/audiowmark
You can stuff per-item database unique IDs, user IDs, geohashes, and other nefarious things inside.
We need to protest this LOUDLY.
Our devices are being locked down, we're having attestation and trusted computing forced on us, the internet all over the world is undergoing age verification with full ID verification.
Just because this is on "ai images" today doesn't mean it won't be on all images - screenshots, your camera reel, etc. - in the fullness of time.
This is scary.
These are the tools of 1984. They've been boiling the water slowly, but in the last year things have really started to pick up pace. Please push back. Loudly.
Everyone at Google and OpenAI working on this: WHAT THE FUCK ARE YOU DOING. STOP.
We have laws and mechanisms to prevent revenge porn, CSAM, defamation, etc. They are robust and can be made even stronger. We do not need to sacrifice the security of our privacy and our speech to fight imagined harms when the real danger is turning into an authoritarian society.
If the powers-that-be want to enforce age verification, watermarking camera output is not the correct technology to do so. It would be something like HDCP, where camera manufacturers are given keys and a whole trusted media path is built so that the relying party can cryptographically enforce that a trusted camera is being used to capture live images.
You can still use traditional methods to manipulate images, too, so I don't think a "does not contain SynthID watermark" means you can trust that image more. In the other hand, encoding a lot of personal and other information in the watermark (136 bit is a lot) that can not be easily removed and most of the people are unaware of it seems really an 1984-like dystopia.
The same techniques used here can be applied in other domains for other purposes. That would not "defeat its only purpose". The danger is the normalization of watermarking for [ insert good reason here ] with regulation eventually making it mandatory once everyone is accustomed to it. Rinse and repeat to gradually boil the frog.
We live in a world where nearly all printers already watermark everything they print with their serial number. It wouldn't be at all surprising if the next modernized variant of that technology encoded personal and contextual data tied to the user.
Zero watermarks is a lot worse than semi-effective AI watermarks.
Watermark, by design, irreversibly modifies the original data, and is, by design, hard to remove without producing detectable artifacts (or rendering the data useless altogether).
In short, the answer is no.
As someone that creates things with tools with different media I would just hard avoid this tool that adds...
arbitrary metadata not of my choosing.
Should I seriously make a texture for a videogame with this weird DRM glorp in it?
How old is photoshop and why is it exempt?
> How old is photoshop and why is it exempt?
For one, it's not developed by Google or OpenAI. The barrier to entry to making realistic but deceptive images with Photoshop is far higher than with AI, and there are already techniques that can, imperfectly, be used to detect the use of traditional image editing.
There is no case that any of its particularly harmful outside of things like CSAM which is illegal.
If this actually works solidly, Google is in deep, deep, deep shit. It would mean that I can put a mark on my non-AI videos and demand that Google not allow upload of my identifiably copyrighted content.
This would completely obliterate YouTube.
I'm sure you can think of a couple things that differentiate gen AI from photoshop, I believe in you.
Its a tool with different modalties and affordances.
But on second thought it is not a bad idea to be able to have a quick tool to identify an image as AI generated.
And after reading your reaction to it, I am sure now that the watermark is for the best.
Only criminals and bad actors want private defaults?
The burden of proof is proving there is some harm or problem that needs solving and noone has managed that in this thread or generally.
No, but you are in the school that teaches that false equivalence is valid rationale.
> Only criminals and bad actors want private defaults
As I was saying.
> The burden of proof is proving there is some harm or problem that needs solving and noone has managed that in this thread or generally.
"Burden of proof" is a concept borrowed from legal practice where the accuser has to offer proof that the accused commited a crime.
No crime is being implied here. Watermarking is actually a useful feature so that people can easily identify images as AI generated.
SynthID would only be DRM if Google/OpenAI were claiming IP rights over their images. I don’t even know if that’s legal though.
So that you don't have to address any of the issues?
https://en.wikipedia.org/wiki/Digital_rights_management#Wate...
> They are not complete DRM mechanisms in their own right, but are used as part of a system for copyright enforcement ...
Because watermarks in and of themselves are not, in fact, DRM. Even if I agree that their mass adoption by BigTech is a really bad sign for personal privacy and (eventually) freedom.
If you read my original point you'd see I said "weird DRM glorp" which you and other have tried, and failed to only closely parse "DRM" so that you could nitpick poorly.
It is integral and part of DRM systems and certainly "weird DRM glorp" for an actual close reader.
DRM is not just "I cant watch X movie because DRM" even if that is the statistically prevalent understanding of DRM.
Its a suite of technologies of which watermarking is one of.
So strictly speaking brings a lot to the discussion when you actually think about it. Stating that DRM != SynthID is addressing issues where people seem to think that DRM == SynthID. Those people are wrong, and strictly speaking need to be corrected.
"this image made by OpenAI" is a drm assertion
You wont be able to assert copyright of the picture that you added an OpenAI red bowtie to, thats a DRM issue.
How does today’s maximum theoretical disinformation output per minute compare to 2021 Photoshop?
So weird images are a big problem? No they don't matter at all.
So what does a deepfake matter?
A national news story in the US tonight, Lyft driver caught faking photos of his messy car. Not the most intelligent fraudster as he left the Gemini logo on the corner of the image.
Providing these four examples in good faith :) also generally I _dislike_ DRM
You should also think about whether, suddenly, courts can now trust images they see because this technology exists?
I think thats not even basically plausible.
What image is going to change your worldview so radically that the drm saves you?
edit - to be clear you are watermarking 100,000 fishes with mustaches because of your concern over 1 image that "matters" (and you don't even have an image that matters in mind)
Also you: well, games go through some kind of distribution, which has plenty of telemetry and metadata. Whether it is App Store with notarization, or Steam or Itch who collect analytics and know a lot about you, or your ISP if you self host your eclectic WebGL game from home. Posting on an iPhone or Android phone, to hacker News which has your email address, on your cell network which has IPv6 globally unique addresses...
"But my choosing!" You'll say. It is extremely performative of you to say, "everything that would make me 200% wrong isn't valid."
I don't know. I really hate these vibes-driven reactions to (checks notes) content attribution. Every accusation is a confession in this frame of mind. How do you not see that?
I have an IP address so therefore this is all fine?
"Every accusation is a confession" also seems like an insinuation that I have something to hide but you have "nothing to hide, nothing to fear"ie the very generic privacy right fallacy.
As for "vibes driven"... this whole technical "fix" is a result of the reactionary "vibe" of the ai moral panic, your "notes" don't seem to be providing any perspective there?
If I take a screenshot of an AI image, will that then be seen as an AI image? Is that 'hidden in the image' or as metadata?
I tested the day 1 when Nano Banana Pro was released and it worked. It still works today for Nano Banana 2.
I didn't post this anywhere because I (arrogantly) thought saying it publicly would make the internet worse. But it was pure arrogancy: if I came up with this the first day then of course other millions of programmers did too.
That being said, it'll introduce the typical artifacts from SD models and that might be detected by other methods (or just by zooming in a lot and looking carefully).
Never released it, but it was obvious to most people in the SD community that denoising using a diffusion model was a relatively trivial means to beat most steganographic watermarks.
Don't sell yourself short. I'm sure it was only hundreds of thousands.
In my tests the image looks clearly distinct. In other words, if you can tell the difference then it isn’t a good test.
Well, they'll finally find out that no one wants to look at AI generated pictures or text. Once they do that, the tool will fail for the public and only work for the government.
If social media platforms started banning images with these watermarks seems like they'd be stripped out overnight.
Set up as a ComfyUI workflow that does a few things: it tries SDXL, Flux, and a couple of different denoising methods at the lowest possible strength (progressively incrementing) to avoid changing the image too much, while also running a SynthID check each time, and repeating this in a loop until the watermark is essentially gone.
At the same time, you’d probably want to add some kind of threshold based on a perceptual hash aka the maximum perceptual quality difference you’re willing to accept.
Writing a more detailed description does not make the models stick to it more.
Comparing Qwen-Image, Flux.2, ZiT, NB2, and gpt-image-2