Google Gemini watermark toggle lets builders hide visible AI markers while keeping invisible provenance

Google now lets you turn off the visible watermark on AI-generated images, videos, and music created with Gemini and Flow. The new Media Watermark toggle removes the "sparkle" icon from content produced by Nano Banana, Omni, and Lyria models. But invisible SynthID watermarks and C2PA metadata stay embedded in the background, so provenance remains verifiable.

What the Google Gemini watermark toggle actually does

The Media Watermark toggle is a user-facing setting in Gemini and Flow that controls whether a visible "sparkle" watermark appears on AI-generated content. When turned off, the watermark is removed from images, videos, and music created with Google's Nano Banana, Omni, and Lyria models. The toggle applies to all future content you generate, and you can switch it back on at any time.

Google VP Josh Woodward confirmed the change, noting that the feature was one of the top user requests. The toggle is rolling out over the coming days to Gemini and Flow, with a planned expansion to Google Search.

What stays invisible: SynthID and C2PA

Even with the visible watermark disabled, Google still embeds invisible SynthID watermarks and C2PA metadata into every piece of AI-generated content. These markers cannot be toggled off. You can verify AI origin by asking Gemini or Google Search to check for SynthID or C2PA markers. This preserves back-end traceability for compliance, forensics, and content moderation.

Why this matters for AI builders

If you build products that ingest or display Gemini-generated media, you can no longer rely on the visible watermark as a quick signal of AI origin. You will need to check SynthID or C2PA programmatically if you need to verify provenance. For products that display AI-generated content, you may want to add your own labeling since the visible marker is now optional.

The change aligns with broader industry trends. OpenAI and Meta already avoid visible watermarks, relying on SynthID and C2PA. Anthropic recently announced invisible watermarks for Claude-generated text and images. Google's move gives creators more flexibility while maintaining a safety net through invisible markers.

Practical implications for your workflow

• If your pipeline assumes visible watermarks are always present, update your content processing logic to handle both cases.
• For compliance with regulations like the EU AI Act, you can still verify AI origin through SynthID or C2PA queries, but this requires an active check rather than a visual scan.
• Region and account restrictions matter. In India, South Korea, and Vietnam, the toggle requires a paid AI Ultra subscription. Work and school accounts do not get the setting at all. The EU and other jurisdictions that mandate visible watermarks by law are excluded.

Caveats and restrictions

The toggle is not available everywhere. Countries that require visible watermarks by law, such as EU member states, will not see the option. Work and school accounts are also excluded. Invisible markers remain, but they require active querying to detect. The feature is rolling out gradually, so availability may vary by app version and platform.