【Dear Utol (2025): WEEK 5 HIGHLIGHTS Episode 29】

It's bad enough that tech companies trained their AI models on Dear Utol (2025): WEEK 5 HIGHLIGHTS Episode 29content that doesn't belong to them. Now, however, it appears that at least one AI model is being used to remove watermarks from images so others can use content that doesn't belong to them, too.

Social media users have recently discovered that Google's new Gemini 2.0 Flash AI model can be used to remove watermarks from images.

This Tweet is currently unavailable. It might be loading or has been removed.

As reported by TechCrunch, this isn't completely new. Other AI image models have been able to remove watermarks. However, Gemini 2.0 Flash appears to be better than all other AI models at this particular task. 

Gemini 2.0 Flash doesn't just remove the watermark. It fills in the gaps in the image that are left from removing the watermark.

Based on what users on platforms like Xand Reddithave shared, it appears that ​​Gemini 2.0 Flash does have some trouble removing certain types of watermarks, such as semi-transparent watermarks. As TechCrunch notes, Google has only made the model available via its developer tools platform, and the company has currently labeled Gemini 2.0 Flash’s image generation feature as “experimental” and “not for production use."

Reddit

However, for photographers and other artists, this is certainly a concerning use of the tool, at least until Google puts some guardrails on the AI image generation feature.

Photographers often depend on watermarks to identify their work as their own; when someone purchases the work, the watermark is removed for that person's use. But if any internet user can just run the image through an AI model to remove the watermark, artists will certainly experience problems getting paid. (Note, also, that in most cases removing a watermark without the permission of the creator is copyright infringement, and illegal under U.S. law.)

The future issues here likely won't be with Google's specific AI tool. The issue is that this is possible with an AI model to begin with. Even if Google adds guardrails to protect copyright holders from uses such as this, there will likely be other third-party AI tools that will replicate these features. In fact, as we previously mentioned, they already are. And soon, they'll probably be just as good as Google's Gemini 2.0 Flash model, too.

(Editor: {typename type="name"/})