The familiar sparkle in the corner of AI-made images is about to become optional — but that doesn't mean the content is untraceable.
For anyone who has generated an image or a video with Google's AI tools, one detail has often been hard to miss: a small, visible watermark stamped into the corner, indicating that AI was involved in the creation.
That is now changing.
Google announced on Friday that users will be able to switch off the visible watermark on eligible content produced by its AI models. The change covers supported images and videos as well as music, generated through models such as Nano Banana, Omni, and Lyria. The toggle is arriving first in the Gemini app and in Flow, Google's AI-powered video creation and editing tool, with support in Search expected to follow.
The update was announced by Josh Woodward, Google's vice president overseeing Gemini, who framed the decision as a trade-off the company has been wrestling with for some time. "We're striking a balance here between creative control and safety," he said.
The reasoning is simple: watermarks get in the way.
A designer dropping AI imagery into a client presentation doesn't want a permanent mark in the corner. Neither does a marketer building a social campaign, or a filmmaker experimenting with generated footage. By making the visible watermark optional, Google is giving creative users more control over how their AI-generated work appears.
Once the feature rolls out to an eligible user, removing the visible mark can be done through the relevant settings. Users can find the new media watermark option and choose whether the visible watermark appears on future generations.
Here's the crucial catch: turning off the visible watermark does not make the content anonymous. Google's SynthID technology continues to provide an invisible signal in supported AI-generated media, helping identify content created or edited with Google's AI systems. Google is also expanding the use of C2PA Content Credentials across supported AI tools to provide information about a piece of content's origin and history.
In practice, that means an image may look completely "clean" to the human eye while still containing underlying provenance information. Google's verification tools can check supported content for signals indicating that Google AI was used to create or edit it.
That distinction matters, however. A verification result indicating no Google SynthID signal does not necessarily prove that content was not generated by AI, because other AI systems use different identification technologies.
The rollout comes with an important geographic asterisk: the visible-watermark option may not be available in jurisdictions where local rules require visible disclosure of AI-generated or synthetic content.
The move has also reignited a familiar debate. Visible watermarks exist partly because they give people an immediate way to recognize synthetic media. Invisible identifiers, by contrast, generally require someone to actively check the content using a verification tool. Critics argue that making visible labels optional makes it easier for AI-generated media to appear indistinguishable from human-created work.
Supporters counter that visible marks were never a foolproof barrier anyway; a determined user could crop or edit them out. Invisible provenance technologies, they argue, provide a more persistent way to identify AI-generated content even when the visible label is absent.
Comments