Google Gemini makes visible AI watermarks optional for users

Google announced on August 14, 2026, that users of Gemini can now disable visible watermarks on AI-generated images, videos, and music. While the visible “sparkle” logo and text can be removed via settings, Google maintains that invisible SynthID markers and C2PA provenance metadata remain mandatory and active for all generated content, according to reports from WhatsNew and Hipertextual.

Settings and compatible Gemini models

Users can toggle the visibility of watermarks by navigating to Settings > Media Watermark within the Gemini app. When disabled, the multicolor star symbol or the phrase “Created with Google AI” will no longer appear superimposed on the media. Hipertextual reports that this setting is not retroactive; images and videos already stored in a user’s library will retain their original watermarks.

The option is initially available for content created with specific models, according to WhatsNew:

• Google Flow: The video editor.
• Lyria: The music generation model.
• Nano Banana: The image generation model.
• Omni: The general generation model.

Support for images generated through Google Search is expected to arrive at a later date.

Technical distinction between visible and invisible marking

According to technical details provided by WhatsNew, Google utilizes two distinct non-visible systems:

SynthID: This system embeds a statistical pattern directly into the pixels of an image. This pattern is imperceptible to the human eye but detectable via Google’s verification API and typically survives basic edits.

C2PA: These are digitally signed provenance metadata attached to the file. While these markers identify the origin of the file, they are removed if a user performs a screenshot or converts the file format.

Industry context and EU AI Act compliance

The move follows a different approach taken by Anthropic, which implemented statistical watermarks on all text generated by Claude on August 2, 2026. This was driven by Article 50 of the EU AI Act, which requires AI providers to ensure outputs are machine-readable for detection. Failure to comply can result in fines of up to 15 million euros or 3% of global turnover, according to WhatsNew.

Google’s decision to make visible logos optional is framed as a distinction between “machine-readable” and “human-visible” markers. The EU AI Act specifies the former; it does not mandate that a logo be visible to the human eye. This aligns Google with competitors like Meta and OpenAI, who do not use visible watermarks by default, according to Hipertextual.

Professional utility versus disinformation risks

Hipertextual reports that the logo hindered the use of AI content in professional creative projects, leading some users to take screenshots to remove the mark—a process that inadvertently stripped the C2PA metadata.

Hipertextual cites a precedent from earlier in the year in Mexico, where AI-generated images of cities in flames were circulated following the reported death of a high-profile drug trafficker. In that instance, some local media outlets shared the content as real, despite the presence of the Gemini watermark.