Google is giving Gemini users the option to remove the visible watermark from AI-generated content, but the company is keeping invisible tracking technology in place to maintain safety and authenticity verification. The new "Media Watermark" setting, rolling out over the next few days in Gemini and Flow with Google Search coming soon, lets users toggle off the sparkle icon that appears on images created with Gemini Nano, videos from Gemini Omni, and songs generated with Lyria.

Why Does Google Allow Watermark Removal Now?

The visible watermark has been a point of friction for Gemini users, particularly when compared to competitors like ChatGPT's latest model, which doesn't display a prominent watermark on generated images. By making the watermark optional, Google is addressing user preferences for cleaner, unmarked outputs while still maintaining a safety net behind the scenes.

The timing reflects a broader tension in the AI industry between creative control and transparency. Users want flexibility in how their generated content looks, but platforms need ways to track and verify AI-generated material as deepfakes and synthetic content become more prevalent.

What Happens to the Invisible Watermarks?

Here's the critical distinction: removing the visible watermark doesn't make the content anonymous or untracked. Google is maintaining two layers of invisible protection that remain embedded regardless of the user's choice:

• SynthID Watermarks: These invisible markers are embedded directly into the pixels of images or the audio data of songs, allowing Gemini to identify whether content was AI-generated when users ask the system to verify it.
• C2PA Metadata: Content Credentials stored in file metadata provide provenance information that can be read by other applications and platforms that support the standard.
• Forensic Detection: Users can still ask Gemini whether an image or video was AI-generated, and the system will provide an accurate answer based on these embedded markers.

According to Josh Woodward, Google's VP for Gemini, the company is "striking a balance between creative control and safety". The approach mirrors Google's broader philosophy: give users aesthetic control while preserving the technical ability to authenticate content.

How to Access the New Watermark Setting

The Media Watermark toggle is straightforward to find and use. Here's how to manage your watermark preferences:

• On Web: Open gemini.google.com, navigate to Settings, find the Media Watermark option, and toggle "Show watermark" on or off.
• On Mobile: Open the Gemini app, access Settings, locate Media Watermark, and adjust the toggle to your preference.
• Scope of Change: Your choice applies to all future content generated through Gemini and Flow, including images from Nano Banana, videos from Omni, and songs from Lyria, though the setting only affects Gemini apps and doesn't strip marks from other Google products.

Who Can Actually Use This Feature?

Google has implemented regional and account-type restrictions that limit access to the watermark toggle. The feature isn't universally available, and some users will continue seeing visible watermarks regardless of their preference settings.

Users in India, South Korea, and Vietnam can only access the watermark toggle if they have an AI Ultra plan; otherwise, the visible watermark remains mandatory. Additionally, work and school accounts never receive the toggle option, meaning institutional users will always see visible watermarks on their generated content. These restrictions reflect varying regulatory requirements and Google's approach to institutional accountability.

What Does This Mean for AI Transparency?

Google's approach differs from other AI companies addressing watermark concerns. Suno, an AI music platform, is embedding audio watermarks so platforms can detect AI-generated songs, while Spotify now flags AI persona artist profiles. Google is taking the opposite direction: hiding the visible mark while keeping the forensic evidence intact.

To support this ecosystem, Google is also open-sourcing Credentio, a C++ library for reading C2PA credentials, allowing other applications and platforms to verify AI-generated content using the same provenance standards. This move could standardize how the internet handles AI content authentication across multiple platforms and services.

The watermark toggle represents a pragmatic compromise in an industry still figuring out how to balance user experience with authenticity verification. As AI-generated content becomes more common, the ability to hide visible marks while maintaining invisible forensic markers may become a standard approach across the industry.