Google Adds Option to Hide Visible Watermarks on Gemini AI Media

Google has introduced a new setting that allows users to toggle off visible watermarks on AI-generated images, videos, and music. The update applies to Gemini Apps and Google's AI video generator, Flow, marking a shift in how the company handles the visual labeling of synthetic media.

Users can now disable the visible "sparkle" mark by navigating to Settings > Media Watermark within the Gemini and Flow interfaces. However, this removal is not universal; Google Support notes that the watermark removal setting is unavailable for work or school accounts. Additionally, in India, South Korea, and Vietnam, the ability to hide these markers is restricted exclusively to AI Ultra subscribers.

The Invisible Layer

Despite the option to remove the surface-level overlay, Google continues to embed invisible SynthID watermarks and C2PA metadata into its generations. According to Google Support, the new setting does not remove these Content Credentials or provenance signals.

SynthID works by embedding imperceptible watermarks directly into pixels and audio waves. This ensures that content can still be identified as AI-generated by specialized tools, even if the visible watermark is absent. This dual-layer approach allows Google to maintain a record of the media's origin without requiring a permanent visual stamp on the final asset.

Balancing Aesthetics and Safety

This move provides significant flexibility for creators who require a professional, clean look for their AI-generated assets. By removing the distracting sparkle overlay, Google is catering to a user base that views visible watermarks as an obstacle to high-quality production and creative integration.

However, the decision to retain invisible tracking reflects the broader industry struggle to combat misinformation. By maintaining C2PA metadata and SynthID, Google attempts to balance user preference for aesthetics with the critical need for AI content traceability. This ensures that while a creator can present a clean image, the underlying data still signals the content's synthetic nature to platforms and verification tools.

What to Watch

As AI-generated media becomes more indistinguishable from reality, the reliance on invisible watermarking will likely increase. The industry is currently watching how effectively these hidden signals survive editing, cropping, or compression. Whether SynthID and C2PA metadata remain robust enough to replace visible labels entirely remains a key point of interest for safety researchers and regulators.