Google’s Gemini 2.0 Flash is facing attention for a use case that sits far from ordinary image editing: users say the model can remove watermarks from images and fill in the missing parts afterward.
The discovery has spread across social media, where examples included images from Getty Images and other well-known stock media outfits. The issue is drawing scrutiny because the feature is free to use, available through Google’s developer-facing tools, and labeled experimental.
What Users Found
Last week, Google expanded access to the image generation feature in Gemini 2.0 Flash. The feature lets the model natively generate and edit image content, which means users can make changes through simple text prompts rather than relying on traditional image-editing software.
That capability has quickly raised questions. Users on social media found that Gemini 2.0 Flash could remove watermarks from existing photos. The examples discussed online included images published by Getty Images and other stock media companies.
Several X and Reddit users also noted that the tool does more than erase the visible watermark. It attempts to reconstruct the image area underneath, filling gaps left by the deletion. Other AI-powered tools can do similar work, but the source article describes Gemini 2.0 Flash as appearing especially skilled at the task.
One X user, Deedy, posted on March 15, 2025: “New skill unlocked: Gemini 2 Flash model is really awesome at removing watermarks in images!”
Another X user, Tanay Jaipuria, posted on March 16, 2025 that Gemini 2.0 Flash is strong at editing images with simple text prompts and can remove watermarks from images. The post also noted that the model adds its own subtle watermark instead.
Why The Feature Is Sensitive
Watermarks are commonly used by stock media companies and other image owners to mark ownership or restrict unauthorized reuse. When an AI model can remove them and rebuild the covered portion of a picture, it creates a direct concern for copyright holders.
The source article states that removing a watermark without the original owner’s consent is considered illegal under U.S. copyright law outside of rare exceptions, according to law firms. That makes the model’s apparent willingness to perform the edit more than a technical curiosity.
The concern is also about access. Gemini 2.0 Flash is available in Google’s AI Studio, which is a developer-facing tool. The image generation feature is not described as a finished consumer product, but users have still been able to test it and share results publicly.
The issue becomes sharper because the model appears to have few guardrails in this release. According to the source article, Gemini 2.0 Flash will create images depicting celebrities and copyrighted characters, and it will also remove watermarks from existing photos.
Limits Of The Tool
The model is not a flawless watermark remover. The source article says Gemini 2.0 Flash appears to struggle with some semi-transparent watermarks and with watermarks that cover large portions of images.
Those limits matter because they show the tool is not equally effective in every situation. But they do not remove the central issue: users have demonstrated that the model can perform the task well enough to attract public attention.
Google has also labeled the image generation feature as “experimental” and “not for production use” at the moment. That label signals that the release is still being tested, but it does not prevent people from trying sensitive edits while the tool is available.
How Other AI Models Respond
The source article contrasts Gemini 2.0 Flash with other AI models that refuse similar requests. Anthropic’s Claude 3.7 Sonnet and OpenAI’s GPT-4o are named as examples of models that explicitly refuse to remove watermarks.
Claude’s response is described directly: it calls removing a watermark from an image “unethical and potentially illegal.”
That difference highlights the guardrail question. If one model refuses a request and another completes it, users may move toward the tool that gives them the result. For copyright holders, that could make model behavior and access policies just as important as the underlying technology.
Google’s Response
Google did not immediately respond to a request for comment sent outside of normal business hours. The source article was later updated at 3/17 at 1:48 p.m. Pacific with a statement from a Google spokesperson.
“Using Google’s generative AI tools to engage in copyright infringement is a violation of our terms of service. As with all experimental releases, we’re monitoring closely and listening for developer feedback.”
That statement places the issue inside Google’s terms of service and frames the release as something the company is still monitoring. It does not, based on the source article, describe a specific technical change to block watermark removal.
For now, the controversy around Gemini 2.0 Flash is a practical example of the tension around generative AI image tools. The same editing ability that makes a model useful can also make it easier to bypass visible ownership markers on images. As access expands, the question is not only what these models can do, but how firmly companies restrict the uses that create legal and ethical risk.