Google Gemini Now Lets You Remove Visible AI Watermarks

Google has added a setting that lets users remove the visible watermark from AI-generated images, videos, and music created in Gemini and its AI video generator Flow. The new "Media watermark" toggle, when switched off, strips the sparkle mark from the bottom-right corner of content produced by Google's Nano Banana and Omni models. Crucially, invisible SynthID watermarks and C2PA metadata stay embedded in the files no matter what the toggle is set to, according to Josh Woodward, VP of Google Labs, Gemini, and AI Studio.

What happened

Detail	Fact
New setting name	Media watermark
Where it appears	Gemini and Flow (Google’s AI video generator)
What it removes	Visible sparkle watermark, bottom-right corner
Affected models	Nano Banana and Omni
Content types covered	Images, videos, and music
What stays embedded	Invisible SynthID watermarks and C2PA metadata
Source	Josh Woodward, VP of Google Labs, Gemini, and AI Studio

Google updated Gemini and Flow with a toggle that controls whether a visible sparkle mark appears on AI-generated content. Switch it off and the mark disappears from the corner of any image, video, or piece of music the models produce. Switch it on and the mark returns. The change is live now.

The visible watermark was previously always-on for content from the Nano Banana and Omni models. Making it optional puts Google in line with other AI image tools that do not brand every output by default.

Why it matters

For most business users, the visible watermark was the main friction point when sharing AI-generated assets in client decks, social posts, or ad creatives. Removing it makes the output cleaner without any extra editing step.

That said, the invisible layer does not go away. SynthID is Google’s machine-readable watermarking system, and C2PA (Coalition for Content Provenance and Authenticity) metadata records the origin of the file in a standardised format that some platforms and tools can read. Anyone inspecting the file can still detect that it was AI-generated, even if the human eye sees nothing.

This is a meaningful distinction for brands running social media advertising or any campaign where regulators or platforms may require AI disclosure. The visible mark is cosmetic; the provenance data is structural.

Does removing the watermark make AI content undetectable?

No. According to Josh Woodward, Google embeds invisible SynthID watermarks and C2PA metadata into every piece of content regardless of the visible watermark setting. Tools that check for AI provenance can still flag the file. What changes is only what a viewer sees on screen.

Our take

This is a sensible product decision dressed up as a trust story. Google gets credit for keeping SynthID and C2PA in place, but the real driver here is competitiveness: most rival tools never showed a visible brand mark on outputs in the first place. The toggle closes that gap.

If you are using Gemini-generated visuals in client work or paid campaigns, turn the toggle off and keep it off. The invisible metadata does the provenance job. The visible sparkle was never doing anything useful for your audience anyway.

One thing worth watching: C2PA support across ad platforms and social networks is still patchy. The metadata is there, but whether Meta, TikTok, or Google’s own ad systems actively read and surface it is a separate question. We cover developments like this regularly in our AI news section, and if you want help thinking through how AI-generated assets fit into your production workflow, our AI integration service is a good place to start.

Source: The Verge · AI