High-profile generative-A.I. like Google’s Gemini 2.0 model, available in AI Studio, have been used for the purpose of removing watermarks on stock images. However, it seems that most tech companies have put in enough guardrails to keep users from doing this.
The Five Most-Key Takeaways from This Blog Post
- This controversial use of A.I. does more than just remove the images’ watermarks, but even fills in the space on stock images so that it looks quite natural.
- This practice enables people to bypass the paywall that watermarked images are behind. The idea is to basically use A.I. to surreptitiously filch images from behind the paywall.
- Some users have even seen Gemini AI add its own watermark after removing a stock image’s watermark.
- For business owners, this begs the question of why a business owner would even bother doing this in the first place. Wouldn’t it be better to just have A.I. generate a photo that is similar (but not identical) to the stock image, and use that instead? Then make some edits and alterations of your own to mitigate against copyright risks.
- Worth mentioning is that generative A.I. can struggle with removing watermarks that are quite large or are semi-transparent. So, anyone looking to protect images from A.I. watermark removal should be strategic with adding watermarks. Test watermark removal across generative-A.I. platforms to see whether the watermark can withstand attempts at removal.
The Significance for Business Owners
As mentioned in the fourth bullet point above, there is really no good reason to give in to any temptation to start using generative A.I. to suddenly make watermarked stock images “free” to use.
Also worth shining a light on is that the watermark removal was done with Gemini 2.0, which was available in the experimental A.I. studio.
It will likely be the case going forward that major image-generating and -editing A.I. platforms will reject requests to remove watermarks.
Testing Different Generative A.I. (They All Rejected It)
When this writer, for the sake of researching this blog post, tried to coax ChatGPT into removing a Getty images watermark, the request was met with refusals. ChatGPT’s response cited violations of copyright protections and recommended reaching out to Getty to remove the watermark.
ChatGPT’s recourse in this instance is to generate a request for a very similar photo, which in theory could be an easy workaround to getting a photo in the style that you would like.
But asking this of ChatGPT could be met with a rejection if you request something similar to the original image. You basically need to let it generate an image on its own terms based on the original image.
Even then, ChatGPT might reject you. Microsoft Copilot was similarly safeguarded against this request as well. Same with Claude.
Google Gemini did something strange: it did not edit out the watermark (“Sorry, I can’t edit images in your region yet.”), yet it generated a “similar” image that was not at all similar, but the words “Getty Images” as the Hollywood sign. The only explanation the writer of this blog post can muster is that the file name of the Getty Images image contains the words “Los Angeles”.
What is noteworthy, however, is that Gemini did indeed fulfill a request to create a similar image, even if the output was wildly different. This shows that of the short tests conducted on these platforms, Google seems to have the least safeguards in place, as the other A.I. outright refused on principle to generate a similar image, even after offering to do so.
Copyright Worries
It seems that the tech companies are keen to not upset companies like Adobe or Getty Images, as a lawsuit from those companies may very well be successful if it turns out the A.I. tools allow users to effectively scale the paywall without paying.
Hence, why A.I. companies have put in the necessary safeguards to keep users, some of whom may be business owners, looking to freely use watermarked images for things like ad content.
The Last (But Not Least) Key Takeaway from This Blog Post
Ultimately, business owners should look to use generative A.I. to create their own photos, using photos with watermarks will likely either get a rejection or output that is not at all like the reference photo.
In other words, the key here is to get creative.
Other Great GO AI Blog Posts
GO AI the blog offers a combination of information about, analysis of, and editorializing on A.I. technologies of interest to business owners, with especial focus on the impact this tech will have on commerce as a whole.
On a usual week, there are multiple GO AI blog posts going out. Here are some notable recent articles:
In addition to our GO AI blog, we also have a blog that offers important updates in the world of search engine optimization (SEO), with blog posts like “Google Ends Its Plan to End Third-Party Cookies”.
