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.
Of course if you need to regenerate the image with an unwatermarked image-generation model to remove it, it more or less still serves its purpose.
I would layer two watermarks and let the public remove the most visible one.
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.
Zero watermarks is a lot worse than semi-effective AI watermarks.
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.
Unfortunately, simple statistics mean that you will get a lot more people than just "1" creating disinformation with AI assistance.
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.
- you identify as an artist of sorts,
- you can't tell AI from human images, and
- you sound absolutely pissed off by this.
- those who can tell that an image was AI generated, instantly, even at thumbnail sizes, and manually drawn over
- those who can NOT, regardless of time or resource allotted. Tends to be perpetually angry(my prejudiced PoV).
What could be it? More/less exposure to older, simplistic examples? Prior exposure to human art? Childhood head traumas? Or something else?
ie. you would fail too.
Anyways, that's besides my point. The point of mine is that, it always turn into all-caps flamewars like this, with no middle ground or third camps, and that this has to be more of a phenomenon than regular disagreements. This isn't bikeshedding. This is Spanish bullfighting centered around a piece of red cloth.
1: https://news.ycombinator.com/item?id=42216694
2: I just asked Gemini "is 60% accuracy over 11k participant for a test statistically significant and why", it said "yes, it is overwhelmingly statistically significant" and "completely off the charts". They said p<0.05 figure would be 50.94%.
Most people put so little effort into prompting them that they come out statistically "average" which makes them blatantly ugly.
Most of them are pretty bad just like 90% of everything is bad. Humans create bad art all day long too.
I gave a specific example of a making a texture for a videogame? How does that change what you actually said to me very specifically despite maybe not reading what I said?
To the extent that I'm upset its at peoples capitulation at this DRM nonsense over being overly reactive to ... internet images which dont matter.
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?
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.
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.
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 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
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.
So this is a big win IMO.
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.
C2PA is basically a signature that serves to prove it came from certain source.
It's useful in case you want to prove you got an image from AI model to someone who doesn't believe you.
It's trivially removable and not useful against people trying to pass off generated images as real.
Eventually it won’t matter when image generation is cheap. But few self-host today and few are willing to pay unsubsidized prices, so the vast majority are using the Gemini, OpenAI, and Midjourney. If all 3 adopted SynthID, only a small fraction would use something else.
This is antithetical to freedom and privacy.
There should be no way for anyone to track down who posted a political meme, anti-religious message, or any other legally protected speech. This will come back to bite us in the ass if we keep building it.
Soon every image or communication we make will be watermarked if we continue to let this shit seep into the commons. Everything from your phone photos, to your screenshots, to your social media posts.
One day soon Republicans or Democrats or whoever doesn't like your freedoms will use this tech to identify you and control you.
There are laws for harms - CSAM, revenge porn, etc. Social media platforms can identify, ban, and report abusers. The framework of the law can take care of the rest.
Our digital footprint should not be tracked and barcoded.
> Social media platforms can identify, ban, and report abusers.
& do but Americans nonetheless argue with troll farms[1] every day & it hurts us
[1] 2013-2023, just one known company https://en.wikipedia.org/wiki/Internet_Research_Agency
These laws need a method to know what is true and what is fake. Good luck with that if you can’t tell if neither images, audio or video are true.
This fakes will pave the way for fascists.
How much freedom and privacy will they allow?
Google or anyone else could start adding those unique tracking watermarks you're concerned about any time they want, regardless of whether they use this AI detection watermark, that to be clear can not track you in any way.
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'm not all that worried about stripping it (I'm sure that's trivial).
The problem that I am worried about is that it can be copied (I'd bet $20 that's trivial, too). People WILL put this on images so that they can be "discredited".
I.e. it doesn't make sense for a purely white or black image, but as you gradually add colors or features, at some point they would want to add a watermark, but based on what? It's an interesting question.
It's certain now that most of the Western world has slid into fascism. Privacy and common decency advocates are all but lost.
I will say this, for everyone celebrating this as something that is "extremely beneficial to the cultural moment",
If I were an adversarial nation-state actor, I might be extremely interested in reverse engineering this and poisoning the well by applying it to real images.
Let's make the world impossible to understand.