Google to Let Users Remove Visible AI Watermarks From Nano Banana, Omni, Lyria Content

Google will soon let users strip visible watermarks from AI-generated images and videos, according to an announcement from Google VP Josh Woodward. A new toggle is rolling out for Nano Banana, Omni, and Lyria — models that currently mark their output with a sparkle icon to signal AI generation. Google says the feature will later expand to Gemini and its video editing tool, Flow.

The change effectively undercuts the stated purpose of visible watermarks, which exist to let people quickly identify AI-generated content without needing extra tools. Woodward described the move as "striking a balance between creative control and safety," acknowledging the tradeoff directly.

What's Changing — and What Isn't

The toggle only affects the visible sparkle overlay. Google says it will not touch two other layers of AI content labeling:

• SynthID, Google's invisible digital watermark embedded in pixel or audio data
• C2PA metadata, an industry-standard provenance record attached to files

Both of these remain intact regardless of the toggle setting. However, neither is accessible to a casual viewer — detecting them requires deliberate action, such as asking a chatbot to check an image or uploading a file to a verification portal. That means the average person scrolling through social media will lose the fastest available signal that content was AI-made, even though technical traceability persists.

Regulatory Carve-Outs

According to Woodward, the toggle will roll out "just about everywhere, except in countries where it's required by law to keep them." That includes the European Union, which recently passed AI transparency rules mandating visible watermarks on "authentic-looking" deepfake content, and South Korea, which enacted similar requirements earlier this year. Users in those jurisdictions will not have the option to disable visible watermarks.

Part of a Broader Industry Shift

Google's announcement lands alongside other recent watermarking moves. Suno introduced new "transparency tools" that flag AI-generated songs when they appear on other platforms. Anthropic separately announced it is adding invisible C2PA-based watermarks to text and files generated by Claude, a move explicitly tied to EU compliance requirements.

No pricing, specific rollout date, or full country list was disclosed by Google beyond the EU and South Korea exemptions.

What this means

This is a policy and product toggle, not a new model release — no weights or checkpoints are changing. The practical effect is that visible AI labeling becomes optional rather than default in most markets, shifting detection burden from casual observers to anyone willing to run content through a verification tool. Expect this to sharpen the divide between regions with binding AI-transparency law (EU, South Korea) and those without, and expect more companies to follow Google and Anthropic in treating invisible, metadata-based watermarking as the compliance floor while making visible labels a matter of user preference rather than product policy.