Google lets Gemini users remove visible watermarks from AI-generated content

The new Gemini setting gives users greater control over visible markings on AI-generated media, while Google retains invisible provenance signals through SynthID and C2PA metadata to preserve transparency and traceability.

Google has begun rolling out a new Gemini feature that allows users to switch off visible watermarks on AI-generated images, videos and music. The change gives creators more control over how generated content appears while retaining underlying mechanisms designed to identify its AI origin.

The option is being introduced gradually and can be accessed through Gemini’s Settings menu under “Media Watermark.” Users can select whether the visible watermark should be displayed on content generated in future sessions.

Google said the feature will be available across Gemini-generated images, videos and music, although visible markings will continue to be shown in jurisdictions where regulations require them.

The move follows feedback from Gemini users seeking greater flexibility over the presentation of AI-created material. Josh Woodward, Google’s vice president overseeing Gemini, AI Studio and Google Labs, said the same functionality is also planned for Flow, with Google Search expected to receive it subsequently.

Invisible provenance markers stay in place

Switching off the visible watermark does not mean that Google-generated content becomes entirely unmarked. The company said its invisible SynthID watermark and C2PA metadata will continue to be incorporated into generated media.

These technologies provide provenance information that can remain associated with a file even when a visible indicator is no longer displayed. Google said the approach is intended to give creators greater control while maintaining mechanisms for identifying the origins of AI-generated content.

Users can also use Gemini or Google Search to check whether an image was created using AI, according to the company.

Google is simultaneously expanding its work around digital content provenance by open-sourcing Credentio, a C++ library designed to work with C2PA Content Credentials.

According to Google’s developer documentation, Credentio enables provenance information to be validated locally, allowing developers to examine credentials without transferring media files to cloud services or external verification systems. The library focuses on validating C2PA credentials, including digital signatures, manifests and embedded claims.

Google said Credentio draws on technology used across nearly 40 of its C2PA-enabled products and has supported billions of digital assets spanning images, videos, audio and documents.

AI industry expands invisible content marking

Google’s approach comes as AI companies increasingly explore machine-readable methods to identify synthetic content without relying solely on visible labels.

Anthropic has introduced a system for embedding invisible, machine-readable markers into text produced by newer Claude models. The company describes the technology as a provenance mechanism rather than a definitive AI detector, noting that its presence may indicate that content has been processed by Claude, while its absence cannot establish that AI was not involved.

Anthropic has linked the initiative to transparency requirements under Article 50(2) of the European Union’s AI Act. The marking system is being applied across supported Claude offerings, including its consumer service, API and coding and enterprise products.

The developments point to an emerging industry model in which visible AI labels can become optional in some circumstances, while hidden provenance technologies continue operating in the background to support transparency and content verification.