Users will be able to manage the feature through a Media watermark section in Settings, where the visible label can be switched on or off. Google is bringing the same control to Flow, its AI video editor, and plans to support it in Search later. The option will not be offered in countries where visible watermarks are legally required, though Google has not identified those markets.
Removing the spark icon does not eliminate Google's mechanisms for identifying AI-generated content. SynthID continues to place an invisible signal in generated media, while C2PA metadata provides another layer of provenance information. Google said SynthID has already been applied to more than 10 billion pieces of content.
Josh Woodward, Google's vice president for Gemini, described the change as an attempt to preserve user control without abandoning AI disclosure mechanisms. “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. So you can still use Gemini or Search to see if an image was AI-generated.”
That approach changes how people may encounter Google's AI disclosures. When the visible watermark is disabled, viewers will no longer be able to rely on the spark icon alone to recognize media generated with Google's tools. Instead, Gemini and Search can inspect the provenance information retained in the file.
Google is also extending that verification capability beyond its own products by open-sourcing Credentio, a C2PA library designed for developers. The software allows applications to validate content credentials locally, giving third-party developers another way to incorporate provenance checks into their products.
The watermark change arrives as Google continues updating Gemini. Gemini 3.7 Flash has replaced 3.6 Flash in the model picker and is being used for chats as well as the Spark agent. The model appears alongside 3.5 Flash-Lite and 3.1 Pro.
Google's decision also comes shortly after Anthropic took a different approach to disclosure. Anthropic added watermarks to text and files generated by Claude, citing compliance with European Union regulations. Some users publicly criticized that decision because the markings could reveal their use of AI in work or academic settings.
The two approaches place different emphasis on what users can see directly. Google is making its visible indicator optional for generated media while retaining machine-readable provenance, whereas Anthropic has introduced markings on types of Claude output that previously did not carry them.
For Google, the core identification infrastructure remains mandatory even as the corner icon becomes a user choice. SynthID and C2PA continue to provide the technical signals behind provenance checks, while Credentio gives developers another route for detecting those credentials outside Google's own services.
This analysis is based on reporting from SQ Magazine.
Image courtesy of Google.
This article was generated with AI assistance and reviewed for accuracy and quality.
