Google lets Gemini users turn off visible AI watermarks on images, video, and music

Google is rolling out a setting that lets Gemini users disable the visible watermark that has stamped every AI-generated image, video, and song, while keeping the invisible tracking layer intact behind the scenes. The toggle went live August 14, 2026, and covers Gemini's image model (Nano Banana), video model (Omni), and music model (Lyria).

What's new

According to TechCrunch, which first reported the rollout: "Google will now allow users to remove a visible watermark from its AI generations, including images, videos, and songs." The control lives under Settings > Media Watermark in the Gemini app, with a simple on/off toggle. Coverage extends beyond the core Gemini app: "The setting to turn the visible watermark off will be available in Gemini and Google's video editor, Flow, with support for Search coming soon."

Critically, the change only affects the visible corner mark, not Google's underlying provenance system. As TechCrunch notes: "While the visible watermarks are now optional, invisible SynthID watermarks and C2PA metadata are still being used for transparency." Every file generated by Gemini still carries an embedded SynthID signature and C2PA metadata regardless of how the visible-watermark toggle is set. The option is also not universal — it will not be available in jurisdictions where local law requires visible AI labeling, and reporting indicates some markets (including India, South Korea, and Vietnam) will require a paid AI Ultra subscription to access it, while work and school accounts don't get the setting at all.

Context

Google introduced the visible spark-icon watermark alongside its invisible SynthID system as part of a broader push, shared across the industry, to make AI-generated media identifiable at a glance. Anthropic took a related step earlier this month, adding invisible watermarking and C2PA metadata to Claude-generated text and files. Visible watermarks in particular have drawn steady user complaints, especially from paying subscribers who felt a corner logo undercut output they were paying to generate, and third parties have already published workarounds for stripping the mark from past Gemini/Veo outputs.

Why it matters

The move draws a sharper line between disclosure and detectability: Google is giving users control over the visible signal while preserving the invisible one, betting that SynthID and C2PA metadata are enough to satisfy provenance and platform-integrity goals even when the on-image branding is gone. That's a meaningful shift for a company that has treated visible watermarking as a default trust signal for AI media. It also puts Google's policy in tension with jurisdictions that mandate visible labeling by law, which is why the toggle won't be available everywhere — a preview of how AI-content-labeling rules are starting to fragment by region rather than converge on one global standard.