Gemini Adds Optional Toggle to Remove AI Visible Watermarks

Introduction

On Friday, August 14, 2026, Josh Woodward, Google's VP for Gemini, announced on X that the visible watermark Google stamps on its AI-generated images, videos, and songs is becoming optional. A new "Media Watermark" setting, an on/off toggle in Gemini's Settings menu, is rolling out over the following days. The change applies to content generated by Google's Nano Banana, Omni, and Lyria models, in the Gemini app and in Flow, Google's AI video editor, with support in Search described as coming soon.

The visible mark, a small logo stamped in the corner of generated media, has been a point of friction for creators who want to use AI-generated assets in finished work without an overlay identifying them as machine-made. Google's answer is to make that visible mark optional while keeping its invisible provenance systems, SynthID watermarking and C2PA metadata, active no matter what a user chooses. Woodward framed the change explicitly as a balance: "We're striking a balance here between creative control and safety: while the visible watermarks are now optional, invisible SynthID watermarks and C2PA metadata are still being used for transparency."

Feature Overview

The Media Watermark Toggle

The new setting appears in Gemini's Settings menu under the name "Media Watermark," with a simple on/off toggle. Turning it off removes the visible logo overlay from images, videos, and songs generated by Nano Banana, Omni, and Lyria, Google's respective image, video, and music generation models. The toggle is rolling out in the Gemini app and in Flow; Woodward's announcement described Search support as coming soon, without a specific date.

What Stays Embedded Regardless of the Toggle

Turning off the visible watermark does not disable Google's other two provenance mechanisms. SynthID, Google DeepMind's invisible watermarking technology, remains embedded in the generated file's underlying data and can still be detected by tools built to read it. C2PA metadata, industry-standard machine-readable Content Credentials that record how and when a file was generated, also stays attached regardless of the toggle setting. Woodward's own statement frames this as the deliberate trade-off behind the feature: visible marks become a matter of user choice, while the two invisible, harder-to-strip layers of provenance remain constant.

Regional Availability Limits

The toggle will not be offered in countries whose local rules require a visible watermark on AI-generated content. Google's announcement did not name specific countries, but the carve-out means the feature's availability varies by jurisdiction based on local regulatory requirements around visible AI-content labeling.

Credentio: A Related, Separate Release

The same week, Google open-sourced Credentio, a C++ library for locally validating C2PA Content Credentials, meaning validation happens on-device rather than by uploading a file to a cloud service. Credentio supports C2PA specification versions 2.2 and 2.4, and supports both official and custom trust lists for determining which signing sources to treat as verified. According to Google, the same underlying validation code has already run across nearly 40 of its C2PA-conformant products, covering tens of billions of generated images, videos, audio files, and documents. As released, Credentio validates existing Content Credentials; it does not yet support creating or embedding new credentials, though Google has said that capability is planned. Credentio is context for how Google's broader provenance stack works alongside the new watermark toggle, not a separate headline feature of this release.

Usability Analysis

For creators who generate images, video, or music with Gemini and want to use that output in finished work, presentations, or commercial projects, the visible watermark has been the most immediate friction point: a small logo in the corner that had to be cropped out or that signaled, at a glance, that content was AI-generated. The new toggle removes that friction directly, and it takes effect the moment a user flips the setting, with no separate approval step described in Google's announcement.

What the toggle does not change is detectability by anyone equipped to check. SynthID and C2PA metadata persist in the file regardless of the visible-watermark setting, so platforms, publishers, or researchers running SynthID detection tools or reading C2PA Content Credentials can still identify the content as AI-generated. The practical gap is between an ordinary viewer, who has no way to run a SynthID detector on a video they encounter online and can no longer rely on a glance at the corner of the frame, and a technically equipped platform or investigator, who can still verify origin through the invisible layers. Credentio, Google's newly open-sourced validator, is aimed at closing part of that gap for developers building detection into their own products, since it runs the same validation logic Google uses internally without requiring a cloud upload. But it validates, it does not create, so ordinary consumers still have no built-in way to check provenance themselves without a tool that implements C2PA reading.

Pros and Cons

Pros:

• Gives creators direct control over whether a visible watermark appears on their generated images, video, and music, addressing a common complaint about needing to crop or edit around the mark
• Preserves invisible provenance (SynthID and C2PA metadata) regardless of the toggle, so the change does not eliminate Google's ability, or a downstream platform's ability, to verify AI origin
• Country-level restriction where local law requires visible watermarking shows the feature was designed with existing disclosure regulation in mind rather than as a blanket global opt-out
• The concurrent open-sourcing of Credentio gives developers a way to build local C2PA validation into their own products using code Google says has already been proven across tens of billions of files

Cons:

• Ordinary viewers who encounter AI-generated media online have no practical way to run a SynthID detector or read C2PA metadata themselves, so removing the visible cue reduces at-a-glance transparency for anyone without specialized tooling
• Search support is described only as "coming soon" with no committed date, so the feature's rollout across Google's own surfaces is incomplete at launch
• Credentio currently only validates C2PA credentials; it cannot yet create or embed them, so the open-source release does not by itself expand who can attach verifiable provenance to their own content

Outlook

The shift from a mandatory visible mark to an optional one puts more weight on invisible provenance systems doing the transparency work that a visible logo previously did automatically for casual viewers. That is a defensible trade-off for creators who found the visible mark impractical, but it depends on SynthID detection and C2PA reading becoming more accessible to platforms, browsers, and eventually ordinary users, not just to specialists. Credentio's release, alongside the toggle, is a step in that direction for developers, since it lowers the bar for building local validation into a product without depending on a cloud-hosted checker. Whether that translates into consumer-facing verification tools, in browsers or social platforms, is a separate question the toggle and the library don't answer on their own.

The regional carve-out for countries with visible-watermark requirements is also worth watching. Provisions like the EU AI Act's Article 50 transparency duties for AI-generated content mean Google's global rollout will likely stay fragmented by jurisdiction for as long as different regulators set different disclosure rules, rather than converging on one global default.

Conclusion

Google's new Media Watermark toggle gives Gemini, Flow, and eventually Search users a real choice about visible AI labeling on their generated images, videos, and songs, while keeping SynthID and C2PA metadata intact underneath regardless of that choice. It is a meaningful concession to creators who found the mandatory visible mark impractical, paired with a clear statement from Google that invisible provenance is the layer doing the actual transparency work going forward. Creators who need to remove a visible mark from finished work get immediate relief; anyone relying on a glance at the corner of an image to know whether it was AI-generated will need to rely on someone else's detector instead.

Editor's Verdict

Gemini Adds Optional Toggle to Remove AI Visible Watermarks earns a solid recommendation within the Gemini space.

The strongest case for paying attention: the toggle gives creators direct control over whether a visible watermark appears on Gemini-generated images, video, and music. That alone raises the bar for what readers should expect in this space. Reinforcing that, it preserves invisible SynthID watermarking and C2PA metadata regardless of the toggle, maintaining a verifiable provenance layer — practical value rather than just headline appeal. The broader signal worth registering is straightforward: making the visible watermark optional shifts the burden of at-a-glance AI transparency onto invisible SynthID and C2PA layers that ordinary viewers cannot check without specialized tools. On the other side of the ledger, one constraint is real rather than a marketing footnote: ordinary viewers without SynthID detection or C2PA-reading tools lose their main at-a-glance way to spot AI-generated content once the visible mark is off. It should factor into any serious decision. Layered on top of that, Search support is only described as "coming soon" with no committed date — which narrows the set of teams for whom this is an obvious yes.

For Google Cloud and Workspace integrators, multimodal-first teams, and Gemini API adopters, this is a serious evaluation candidate, not just a curiosity to bookmark. For everyone else, the safer posture is to monitor coverage and revisit once the use cases that matter to your team are demonstrated in the wild.