Google Lets Users Disable Visible Watermarks on Gemini-Generated Media

Google is giving Gemini and Flow users the option to remove the visible watermark stamped on AI-generated images, videos, and music. A new "Media watermark" setting in Gemini's settings menu lets users toggle off the "sparkle" icon that normally appears in the bottom-right corner of content produced by Google's Nano Banana and Omni models.

The change does not eliminate Google's watermarking system entirely. According to Josh Woodward, vice president of Google Labs, Gemini, and AI Studio, content generated through Google's tools will still carry invisible SynthID watermarks and C2PA metadata embedded in the file, even when the visible marker is switched off. Users can check whether a piece of content carries these markers by asking Gemini or Google Search directly.

Google plans to extend the watermark toggle to Search as well. The company said the setting will not be available in countries where regulations require visible watermarks on AI-generated content, though it did not specify which countries fall under that requirement.

Why now

The move brings Google's approach closer in line with competitors. OpenAI does not apply visible watermarks to its image outputs, relying instead on SynthID-style invisible signals and C2PA metadata. Meta has built its own invisible labeling system, called Content Seal, rather than stamping a visible mark on generated media. Anthropic this week said it is now applying invisible watermarks to both AI-generated text and images, expanding the industry's shift toward metadata-based provenance over visible markers.

Google's own framing is that visible watermarks were becoming a weak deterrent anyway. Many users have already been cropping or editing out the sparkle icon before sharing AI-generated content, undercutting its purpose as a quick visual signal. By making the setting official rather than leaving removal to third-party editing, Google is acknowledging that visible watermarks function more as a default courtesy than a hard technical safeguard.

What this means

This is a policy and settings change, not a new model release — it affects how existing Nano Banana and Omni outputs are labeled, not the models themselves. The practical effect is that casual viewers scrolling through social feeds or messaging apps will find it harder to tell AI-generated media apart from real photos and video at a glance, since the obvious visual cue becomes optional.

The invisible layer — SynthID and C2PA — becomes the real line of defense for verification, but it only works if someone thinks to check. Journalists, platforms, and content moderators will need to actively query tools like Gemini or Search rather than relying on a glance. As every major lab (Google, OpenAI, Meta, Anthropic) converges on invisible-only watermarking as the default, the burden of detecting synthetic media shifts further onto automated verification systems and platform-level scanning, rather than resting on something a human can see immediately.