Google Lets Users Turn Off Visible Watermarks on Gemini AI Generations
The invisible SynthID signature and C2PA provenance metadata stay embedded either way — only the visible on-image watermark becomes optional, and the rollout is gradual and geography-gated.
By Dillip Chowdary • Aug 14, 2026 • Source: TechCrunch
Google announced it will let users remove the visible watermark from Gemini's AI generations — images, video, and songs. The toggle lives under Settings > Media Watermark and covers the Nano Banana, Omni, and Lyria models in Gemini, plus Google's video editor Flow, with Search support coming later.
The invisible watermark isn't part of the deal: SynthID's steganographic signature and C2PA provenance metadata remain embedded in every generation regardless of the visible-watermark setting, and there's no user-facing way to turn either of those off.
That distinction matters for how this actually gets used — commercial designers and production teams who need clean visual outputs get what they wanted, while content-provenance tracking (the part relevant to misinformation and detection tooling) is unaffected.
The rollout started gradually on August 14, 2026, so it may take days to reach individual accounts. In India, South Korea, and Vietnam, the option is currently limited to Google AI Ultra subscribers — other plans in those markets keep the visible watermark on by default.
The move lands the same week Anthropic detailed its own Claude text-watermarking rollout, underscoring that invisible, hard-to-strip provenance signals (not visible marks) are becoming the industry's actual answer to AI-content transparency requirements.