On August 15, 2026, The Verge reported that Google officially launched a high-profile new feature for Gemini and related generative tools: users can choose to turn off explicit (visible) watermarks on AI-generated images, video and music. The update was first announced on social platforms by Gemini lead Josh Woodward and aims to balance creators' creative freedom with the need for content transparency. The feature will initially roll out on Gemini's web interface and Google's video editing tool Flow, and later expand to Search services.

Feature details

The new setting, called “Media Watermark,” can be toggled on or off directly in Gemini's settings. When turned off, content generated by models such as Nano Banana (images), Omni (video) and Lyria (music) will no longer display the visible “sparkle” watermark in the lower-right corner. The rollout is being phased; in some regions (for example India, South Korea and Vietnam) the option may be limited to paid subscribers or legally required visible watermarks may remain in place.

Google stressed that turning off the visible watermark does not mean abandoning provenance protections. Underlying invisible SynthID marks and C2PA-standard metadata will remain intact. Users can still verify whether content was AI-generated via Gemini or Google Search tools. Woodward said this represents a compromise “between creative control and safety.”

At the same time, Google open-sourced a C++ local verification development library called Credentio. The library supports C2PA Content Credentials (specs 2.2 and 2.4) and enables high-performance verification locally without uploading media files to the cloud, addressing privacy, latency and scalability concerns. Credentio already supports nearly 40 internal Google products that comply with the C2PA standard and has processed billions of generated assets.

Technical and policy background

In recent years, visible watermarks on AI-generated content have been a frequent user complaint because they affect professional aesthetics and post-production workflows. Many creators have had to remove watermarks manually, adding to their workload. Google's adjustment directly responds to this pain point while maintaining transparency through invisible marks and standardized metadata, aligning with global regulatory trends on AI content provenance.

SynthID is an invisible watermarking technology developed by Google DeepMind that embeds detectable signals without noticeably affecting perceived quality; C2PA is a content authenticity standard driven by Adobe, Microsoft, Intel and others that records generation and editing histories via cryptographic signatures.

Potential impacts: improved creative efficiency and ongoing provenance contest

1. Impact on creators and general users — Making the visible watermark optional will significantly reduce post-processing steps, especially benefiting professional scenarios like design, marketing and video production. Users can more freely integrate AI-generated content into workflows, improving efficiency and final output quality. The availability of verification tools means content provenance can still be queried.

2. Impact on Google's ecosystem — The feature rollout should help boost Gemini's competitiveness among creative tools and attract more professional users. Open-sourcing Credentio further strengthens Google's influence in content provenance standards and encourages third parties to integrate local verification capabilities.

3. Impact on the industry and regulators — The move may prompt other AI vendors to adopt a similar layered watermark strategy of “optional visible, mandatory invisible.” Regulators may focus more on the effectiveness and cross-platform interoperability of invisible watermarks. At the same time, the existence of open-s