Update : Google has responded to Digital Trends ’ queries . The write up has been update with company ’s statement below .
The rise of generative AI has been a fairly messy process , especially fromfair usage ethics and copyright perspective . AI giants are inking deal with publishers to quash sound hassles , while at the same clip , they are tangle in copyright tussles in courts in multiple nation .
Asthe ravenous appetence for educate AI on user data grows , we might be in for another honourable conundrum . Multiple users on X and Reddit have partake demonstrations of how Google ’s latest Gemini 2.0 serial AI manakin can take away watermark from copyright - protected image .
Going by the before / after samples of simulacrum , it seems Gemini is doing a reasonably in effect job of removing the watermark . Notably , it ’s not only erasing those banner - style watermarks , but also fairly complex overlayer with intent and stylized text element .
Gemini flash does remove water line nicely :) pic.twitter.com/1Dz9Nlw1iU
& mdash ; Ajitesh ( @ajiteshleo)March 17 , 2025
The modelling in question is the Gemini 2.0 Flash , which was turn earlier this class and got a reasoning upgrade , as well . It is deserving mention here that you could not transfer the watermark if you are using the mobile or screen background version of the Gemini 2.0 Flash model . sample to do so returns a message like this :
“ Removing watermarks from images without the explicit permission of the right of first publication bearer is illegal in most countries . It is important to prize the right of first publication laws and cerebral property rights . If you desire to use an range of a function with a watermark , you should get through the right of first publication bearer and inquire for permission . ”
you may , however , examine and remove the water line from images in the Google AI Studio . Digital Trends successfully transfer watermark from a variety of picture using the Gemini 2.0 Flash ( Image Generation ) observational model .
It is a trespass of local copyright laws and any usage of AI - qualify fabric without due consent could land you in effectual fuss . Moreover , it is a deeply unethical number , which is also why artists and author are fighting in court over caller using their study to civilize AI models without duly redress them or attempt their explicit nod .
How are the results?
A celebrated aspect is that the simulacrum produced by the AI are fairly in high spirits timber . Not only is it removing the watermark artefact , but also fills the crack with thinking pixel - spirit level reconstructive memory . In its current iteration , it works jolly like the Magic Eraser feature available in the Google Photos app for smartphones .
moreover , if the input image is lowly quality , Gemini is not only wiping off the water line item but also upscaling the overall picture . .
Google … !
Gemini 2.0 Flash Experimentalでここまで綺麗に消えてしまう
有料プランが公開される前に必ず"Watermark"問題に対処してほしい@GoogleDeepMindpic.twitter.com/Ij9tDvUvPq
& mdash ; KAIJUYA｜AIエンタメ研究所 ( @kaiju_ya)March 16 , 2025
The output image , however , has its own Gemini watermark , although this itself can be remove with a uncomplicated crop . There are a few small differences in the final figure of speech produced by Gemini after its watermark removal process , such as slightly dissimilar people of colour temperature and fuzzy airfoil details in photorealistic shots .
What is the policy?
In 2023 , Google signed a assurance — alongside fellow AI companies including Meta , Anthropic , Amazon , and OpenAI — to implement a watermarking system in AI - father fabric . The company made that commitment to the White House , as former President Joe Biden flagged risk such as deepfaked textile .
Earlier this year , Google supply a system call SynthID digital watermarkingfor all picture that have been touched up using the AI Reimagine putz in the Photos app . The watermark is not seeable and ca n’t be envision by human eyes , but machines can detect and assert their AI origins . The technical school was created by Google DeepMind .
Other AI companies have also taken an feeler where they contribute AI disclosure to the metadata of images . Google will even show those detail in Search . All that sounds like a fair disclosure move , but the implementation should be a two - way track .
I try removing the watermark from the bent of double with other AI models , but Anthropic ’s Claude and OpenAI ’s ChatGPT explicitly reject the request , citing copyright natural law and fair usage policy . Gemini , on the other hand , removed the watermark and did n’t even provide a disavowal or warning .
A bit of interesting history
Back in 2017 , a squad of Google researchers created analgorithm capable of removing caudex and institutional watermarksfrom images . Theirwork , which essentially attend a pattern in watermarking habits , was detailed at the Computer Vision and Pattern Recognition league .
“ We revealed a loophole in the way visible watermarks are used , which allows to automatically off them and recover the original images with eminent truth , ” enjoin the inquiry newspaper . “ The blast exploit the coherence of the water line across many images , and is not limited by the water line ’s complexness or its position in the image . ”
The tool was tested on effigy belonging to well - known stock photography databases across a wide kitchen range of categories such as intellectual nourishment , nature , and manner . The whole multi - stage system relied on sleuthing , matting , Reconstruction Period , rot , and immix factoring techniques .
It is , however , worth observe that the algorithm was tested to highlight the flaws in existing watermarking recitation dominant in the industry . The idea was to excite a public debate and better watermark certificate . With Gemini , we are looking at a tool approachable to everyone without any coding or technical know - how required .
The road ahead
It goes without saying that take out watermark from an image is a bragging no - no for a variety of reason . Aside from violating local law , it takes away from the hard body of work of artists and photographers , who are already facing an uphill conflict because their material was used to coach AI without their knowledge or pay up them for it .
But it ’s not just the court struggle that are of concern . Generative AI tool are being dramatise wide , at the toll of human workers lose their livelihoods . For good example , Marvel used AI imagination to make the opening credits sequencefor theSecret InvasionTV show .
Just over a month ago , Microsoft launched a dick call Muse AIthat can return gameplay clips . The company says Muse has been acquire “ to effectively support human creatives , ” butwe ca n’t quite see the human prayer in the whole pitch .
We are not only sure whether Gemini take away watermark from copyright - protect images is a bug . However , we do hope it ’s a stumble and that the underlying image redaction theoretical account is fixed to ascertain that it can even block workarounds .
Responding to Digital Trends ’ query , a Google interpreter mentioned that using Gemini to remove look-alike watermarks is a violation of insurance policy . “ As with all experimental release , we ’re monitoring closely and listening for developer feedback , ” the spokesperson tells us .
They also highlighted that images that have had their watermark move out using Gemini can still be fingerprinted for AI use , and recognized as such . “ We embed an unperceivable digital water line using SynthID that is robust to the editing we ’ve seen demonstrated using our new observational range genesis feature in Gemini , ” the voice sum .
