Google is updating its AI tools to give users more control over how AI-generated content is labeled. Starting now, users can turn off the visible watermark that appears on images, videos, and music created with Google Gemini and the company's AI video generator, Flow. The new setting, called "Media watermark," is available in the settings menu of both platforms.
When the setting is toggled off, Google will remove the "sparkle" watermark that normally appears in the bottom-right corner of content generated by the company's Nano Banana and Omni models. These models power a range of generative features inside Gemini, including image creation, video generation, and music synthesis. For users who prefer cleaner, more seamless outputs, the removal of the visible mark is a significant quality-of-life improvement.
Despite the change, Google is not abandoning AI provenance entirely. The company will continue embedding invisible SynthID watermarks and C2PA metadata into all AI-generated content. According to Josh Woodward, vice president of Google Labs, Gemini, and AI Studio, these hidden markers allow the company and third parties to verify whether a piece of content was produced by AI. Users can ask Gemini or Google Search whether a particular image, video, or audio clip carries one of these markers, enabling a layer of transparency even when no visible label is present.
The decision reflects a growing tension in the AI industry between user experience and content transparency. Visible watermarks have long served as an immediate, human-readable signal that a piece of media is synthetic. They are easy to spot and hard to miss, but they can also be seen as intrusive, especially for creators who want to use AI-generated assets in professional projects. Removing them makes the output look more native, but it also raises questions about how easily synthetic content can be mistaken for real footage or photography.
The rise of invisible provenance technology
SynthID, developed by Google DeepMind, is a technique that embeds an imperceptible digital signature directly into the pixels of an image or the waveform of audio. Unlike a visible stamp, SynthID is designed to survive compression, cropping, and other common edits. It is not meant to be detected by the human eye but rather by software tools that can scan for the hidden marker. This makes it a more robust and durable method of attribution than a simple overlay.
C2PA, which stands for Coalition for Content Provenance and Authenticity, is a broader technical standard that ties metadata to the content's origin and editing history. C2PA metadata can include information about when and where a piece of content was created, what software was used, and whether any edits were made. The standard is supported by a wide range of companies, including Adobe, Microsoft, and Intel, as well as camera manufacturers and news organizations. Embedding C2PA metadata is seen as a way to create a verifiable chain of custody for digital media.
Google's use of both technologies means that even with visible watermarks removed, AI-generated content carries a digital fingerprint that can be used to trace its origin. The company has stated that these markers are resilient to common tampering attempts, making it difficult for someone to strip them out and claim the content as their own. However, no technology is foolproof, and experts continue to study the limits of invisible watermarking against sophisticated attacks.
An industry shift away from visible labels
Google's move places it in line with other major AI companies that have already abandoned visible watermarks. OpenAI, for instance, does not apply a visible badge to images generated by DALL-E or ChatGPT. Instead, it relies on SynthID and C2PA metadata to signal AI provenance. Meta, which offers AI image generation across its platforms, introduced its own Content Seal standard rather than placing a visual watermark on every output. Anthropic announced earlier this week that it is embedding invisible watermarks into both AI-generated text and images from its Claude models.
The industry trend suggests a consensus that visible watermarks are not a sustainable long-term solution. They can be cropped out, painted over, or simply ignored. For years, users have been editing out visible watermarks from AI-generated content using image editors and other tools, which makes the watermark an unreliable signal. By removing the visible label by default, Google and its peers are acknowledging that the future of AI content attribution lies in invisible and embedded technologies rather than cosmetic overlays.
But the shift also introduces new challenges for everyday users and platforms. A person scrolling through social media may no longer instantly recognize an AI-generated image as synthetic. This could make it easier for misleading or fabricated content to spread. Governments and regulators have taken notice, and some jurisdictions have passed laws requiring explicit labeling of AI-generated media. That is one reason Google says the new Media watermark setting will not be available in countries that require visible watermarks by law. In those regions, the toggle will remain disabled, ensuring compliance with local regulations.
How the new setting works
To turn off the visible watermark, users must navigate to Gemini's settings menu and locate the Media watermark option. The same setting appears in Flow, Google's AI video generation tool. According to Google, the toggle applies across the relevant models and content types, including images, videos, and music. When the setting is on, the sparkle watermark appears in the bottom-right corner of the generated content. When it is off, no visible mark is rendered.
The change does not affect existing content that has already been generated. It applies only to new outputs created after the setting is modified. Users who want to keep visible watermarks on their AI-generated content for clarity or professional reasons can simply leave the setting enabled. The choice is entirely up to the user, which is a departure from Google's earlier approach where the visible watermark was always present.
Google has also indicated that it plans to roll out the watermark setting to Search. This would allow users to control whether AI-generated images surfaced in Google Search results carry a visible label. However, the timeline for that rollout has not been specified. The company has said the Search integration will follow the same guidelines, with the toggle unavailable in jurisdictions that mandate visible watermarks.
Implications for creators and researchers
For digital artists, marketers, and content creators, the ability to remove visible watermarks is a welcome change. AI-generated visuals are often used as placeholders, concept art, or background elements in larger projects. A visible watermark can interfere with those use cases, forcing creators to crop or edit the image before using it. Now, they can generate clean assets directly from Gemini or Flow, saving time and reducing friction.
However, the change also creates new responsibilities for content platforms. Social media networks, news outlets, and video sharing services will need to rely more heavily on invisible metadata and detection tools to identify synthetic content. This could lead to more sophisticated backend filtering and moderation systems. It may also prompt calls for clearer disclosure rules, especially in contexts where AI-generated content could be mistaken for real-world events.
Researchers studying misinformation have expressed mixed feelings about the move. On one hand, invisible watermarks provide a more reliable way to track content performance and provenance. On the other hand, there is concern that the general public will not know how to check for these hidden markers. Without a visible cue, many people may assume that an image or video is real simply because it does not carry an obvious AI label.
The effectiveness of invisible watermarking also depends on widespread adoption of detection tools. Google has built detection capabilities into Gemini and Search, but those tools are only useful if people know they exist and choose to use them. Other companies and third-party developers are working on independent verification services, but there is no universal standard for user-friendly AI content checking yet.
The future of responsible AI labeling
Google's announcement marks another step in the evolving debate over how AI companies should disclose synthetic content. The move away from visible watermarks does not mean the company is abandoning its responsibility to transparency. Rather, it signals a belief that invisible provenance mechanisms are more effective and less intrusive than a permanent visual stamp. By combining SynthID, C2PA metadata, and a user-configurable visible watermark setting, Google is attempting to offer both flexibility and accountability.
The company's decision to keep visible watermarks in countries that require them highlights the importance of legal and regulatory frameworks. As more governments introduce rules around AI content labeling, technology companies will need to adapt their products to meet local requirements. Google's approach allows it to comply with those laws while still offering a watermark-free experience in other regions.
In the short term, users will likely appreciate the cleaner output from Gemini and Flow. In the long term, the success of this approach will depend on how well invisible watermarking technologies hold up and how effectively the public learns to verify AI-generated content. The infrastructure for content provenance is still young, but Google's latest update suggests that the industry is moving toward a future where AI labels are hidden in plain sight, always present but never visible.
Source:The Verge News