Google has introduced an option allowing users to remove visible watermarks from content generated by its AI tools, including Gemini and the Flow video editor. This change provides creators with cleaner files for professional and personal use, while the company maintains underlying provenance tracking through invisible watermarks and metadata.
The newly available "Media watermark" setting permits users to toggle off the visible "sparkle" icon that previously appeared on AI-generated images, videos, and music. This feature applies to content created with Google's Nano Banana, Omni, and Lyria models. The option is rolling out across Gemini and Flow, with integration into Gemini's Search functionality expected soon.
Despite the removal of visible watermarks, Google emphasizes that invisible SynthID watermarks and C2PA metadata will remain embedded in the generated media. These technologies are designed to track the origin of AI-generated content without being perceptible to the human eye. Users can still verify the AI origin of media by submitting it to Gemini or Google Search.
The decision to make visible watermarks optional aims to balance creative control with transparency and safety. Google acknowledges that visible watermarks could be a hindrance for creators needing to use AI-generated content in professional contexts, such as presentations or social media posts, without overt branding. By offering an official way to remove these marks, the company seeks to provide a cleaner output for users, rather than encouraging workarounds.
However, the availability of the visible watermark toggle is not universal. Google has stated that the option will be withheld in regions where local laws mandate visible AI labeling. This includes the European Union, which has implemented regulations requiring clear identification of synthetic media, and South Korea. In some countries, such as South Korea, India, and Vietnam, access to this tool may require a Google Ultra subscription.
The move comes as governments worldwide are increasing scrutiny on AI-generated content and implementing transparency frameworks. The European Union's AI Act is one example of regulatory direction in this area. Google has stated its participation in the EU's AI Act Code of Practice on transparency for AI-generated content and its support for interoperable provenance technologies like C2PA and SynthID.
Google's approach contrasts with that of some other AI companies. For instance, Anthropic recently introduced invisible watermarks for text generated by its Claude model, a move that has also been linked to compliance with EU regulations.
The company has also open-sourced Credentio, a C library designed to allow developers to build local C2PA validation directly into their applications. This initiative aims to facilitate the verification of provenance metadata within third-party apps, rather than relying solely on Google's tools. Google's stated policy on AI-assisted content, documented since February 2023, emphasizes that the quality of content is paramount, not the method of production, unless the primary purpose is to manipulate search rankings.