- Google’s visible-watermark toggle covers images, videos and songs from Nano Banana, Omni and Lyria, with rollout beginning in the coming days. [1] - The change does not remove invisible SynthID signals or C2PA provenance metadata, according to Google’s product announcement. [1] - Google’s own documentation says SynthID detection is probabilistic and can fail after substantial transformations; independent researchers have identified additional removal and false-positive risks.
[2][3] Google will soon let users turn off the visible watermark placed on some AI-generated images, videos and songs, giving creators a cleaner output while leaving the company’s invisible provenance systems enabled. [1]
The setting will initially appear in Gemini and Google’s Flow video editor, with support for Search planned later. It will cover the Nano Banana, Omni and Lyria models and will be available under Settings > Media Watermark as the feature rolls out over the coming days. [1]
What remains in the file #
Google says disabling the visible mark will not affect SynthID, an imperceptible watermark embedded in generated media, or C2PA Content Credentials that record provenance information. Gemini’s verification tools can check images, video and audio for Google’s SynthID signal, although Google says the detector currently identifies content made by Google AI tools rather than all AI systems. [1][2]
The company’s documentation describes SynthID detection as probabilistic. It says the signal usually survives resizing, recoloring, compression and some edits, but may become undetectable after many alterations. For text, Google says confidence can fall sharply after thorough rewriting or translation. [2] Independent researchers at ETH Zurich’s SRI Lab have also studied watermark spoofing and removal attacks against SynthID-Text, while a 2026 study reported high failure rates for several tested watermarking systems after paraphrasing.
[3] [4]Google’s rationale is that visible marks can interfere with creative and commercial use, while invisible markers preserve a way for platforms and users to investigate provenance. That distinction leaves the visible label optional for viewers, even though the underlying file may still carry signals that specialized tools can inspect. [1][2]
Companies mentioned #
Further sources #
[1] TechCrunch reports Google’s August 14, 2026 announcement covering visible-water… ↗
[[2] Google’s Gemini documentation explains SynthID, Content Credentials, supported … ↗](https://support.google.com/gemini/answer/16722517?hl=en)
[[3] ETH Zurich’s SRI Lab describes research into detecting, spoofing and removing G… ↗](https://www.sri.inf.ethz.ch/blog/probingsynthid)
[[4] A July 2026 arXiv study evaluates watermark robustness after paraphrasing and r… ↗](https://arxiv.org/abs/2607.16010)
The stories that matter, in one email. Free — unsubscribe anytime.