Google has introduced a new option that allows users to remove visible watermarks from AI-generated images, videos and songs created with some of its generative AI tools. The change gives creators more control over the appearance of their AI-generated content while retaining Google’s invisible systems for identifying and verifying AI-generated media.
The company said the visible watermark can now be switched off for content generated by models including Nano Banana, Omni and Lyria. The setting is being introduced through Gemini and Google’s Flow video-generation and editing platform, with support for Search expected later. Importantly, removing the visible watermark does not remove Google’s invisible SynthID watermark or C2PA provenance metadata, meaning the content can still be identified as AI-generated through supported verification systems.
Google Makes Visible AI Watermarks Optional
Google previously placed a visible watermark on AI-generated content created through several of its generative tools.
The company is now allowing users to turn off that visible identifier.
The move is aimed at balancing transparency with the needs of creators who may prefer to publish or use AI-generated media without a prominent Google-generated mark appearing on the final output.
| Feature | Details |
|---|---|
| Visible watermark | Can be turned off |
| AI-generated images | Supported |
| AI-generated videos | Supported |
| AI-generated songs | Supported |
| Image model | Nano Banana |
| Video model | Omni |
| Music model | Lyria |
| Gemini support | Rolling out |
| Flow support | Rolling out |
| Google Search support | Planned |
| Invisible watermark | SynthID remains |
| Provenance metadata | C2PA remains |
The change applies to the visible watermark, not Google’s underlying AI-content identification technologies.
What Happens When the Visible Watermark Is Removed?
The most important distinction is between Google’s visible watermark and its invisible provenance technologies.
The visible watermark is designed to provide an immediate visual indication that a piece of content was created with AI.
SynthID works differently. It embeds an imperceptible signal into AI-generated content that can later be detected by compatible tools.
Two Layers of AI Identification
Visible watermark
↓
Immediately visible to viewers
↓
Can now be switched off
+
Invisible SynthID watermark
↓
Not visible to viewers
↓
Remains embedded
+
C2PA metadata
↓
Provides content provenance information
↓
Remains available
This means Google is not abandoning AI-content identification by making the visible watermark optional.
SynthID Will Remain in AI-Generated Content
Google said removing the visible watermark will not remove SynthID.
SynthID is Google’s technology for identifying AI-generated content through an invisible marker.
The system is designed to help determine whether content was generated or modified using Google’s AI systems.
Google’s Approach
AI generation
↓
Visible watermark
↓
Optional
BUT
↓
SynthID
↓
Still embedded
AND
↓
C2PA metadata
↓
Still retained
This allows Google to separate the visual presentation of content from the underlying provenance system.
C2PA Metadata Will Also Remain
Google is also retaining C2PA-related metadata in generated content.
C2PA is an industry standard designed to provide information about the origin and history of digital content.
Together, C2PA metadata and invisible watermarking provide additional mechanisms for determining where AI-generated content came from.
AI Content Provenance
AI-generated media
↓
SynthID
+
C2PA metadata
↓
Machine-readable provenance
↓
Verification
↓
Greater transparency
The approach is becoming increasingly important as AI-generated images, videos and audio become harder to distinguish from traditionally created media.
Nano Banana Content Can Use the New Setting
The visible-watermark change applies to Google’s Nano Banana image-generation technology.
Nano Banana has become one of Google’s prominent image-generation systems and is used within Google’s AI ecosystem.
Users generating images through supported Google tools will be able to control whether the visible watermark appears.
Nano Banana Workflow
Prompt
↓
Nano Banana
↓
AI-generated image
↓
Visible watermark
↓
User can turn it off
↓
SynthID remains
↓
C2PA information remains
This provides greater flexibility for people using AI-generated images for creative and professional purposes.
The Change Also Applies to AI Video
Google is extending the option to AI-generated video created with its Omni model.
Video-generation workflows can be particularly affected by visible watermarks because they may appear across an otherwise finished visual asset.
The ability to switch off the visible mark could make AI-generated video more practical for certain creative workflows.
AI Video Workflow
Prompt
↓
Omni
↓
AI-generated video
↓
Visible watermark
↓
Optional removal
↓
Invisible identification remains
The change could be relevant to creators producing advertisements, social-media content, concept videos and other visual material.
AI-Generated Music Is Included
Google’s Lyria model is also covered by the change.
This means the policy is not limited to images and video.
Users generating songs and other forms of AI-assisted audio can also benefit from greater control over visible AI labeling where the feature is available.
Google’s Generative Media
Images
↓
Nano Banana
Videos
↓
Omni
Music
↓
Lyria
↓
Visible watermark can be disabled
↓
Invisible provenance remains
The broader rollout demonstrates Google’s attempt to create a consistent approach across different types of generative media.
Gemini Will Get the New Setting
The visible-watermark control is being rolled out through Gemini.
Google users will be able to access the relevant media watermark setting where the feature is available.
The rollout may not reach all users and regions simultaneously.
Gemini Workflow
Gemini
↓
Generate media
↓
Media watermark setting
↓
Visible watermark ON/OFF
↓
Generate final content
The exact availability can depend on Google’s rollout and local regulatory requirements.
Flow Will Also Support the Feature
Google’s Flow platform, which focuses on AI-powered video creation and editing, is also included.
This is particularly relevant for professional and creative users who may want to use AI-generated clips as part of larger production workflows.
Flow Production Workflow
Creative concept
↓
AI generation
↓
Video editing
↓
Watermark setting
↓
Final output
↓
Invisible provenance remains
The option could make AI-generated assets easier to integrate into finished projects.
Google Search Support Is Coming
Google also plans to bring the visible-watermark control to Search.
This could eventually allow users generating AI media through Google’s search-based experiences to receive the same degree of control.
The expansion would make the feature available across a broader portion of Google’s AI ecosystem.
Why Google Is Removing the Visible Mark
Visible watermarks provide an obvious signal that an image, video or song was generated by AI.
However, they can also affect the appearance and usability of the content.
For professional creators, a visible Google mark may not fit naturally into an advertisement, presentation, design project or other finished work.
Creator Trade-Off
Visible watermark
↓
Better immediate transparency
BUT
↓
Changes appearance
↓
Can interfere with creative use
↓
Google makes it optional
↓
Invisible identification remains
The company is therefore separating transparency from the visual presentation of the content.
Creators Get More Control
The change could be particularly useful for creators who use generative AI as part of professional workflows.
For example, a designer may want to incorporate an AI-generated image into a larger composition without having a visible watermark in the corner.
Similarly, a video creator may want AI-generated footage to appear consistently with other footage.
Creative Workflow
AI-generated asset
↓
Edit
↓
Combine with other content
↓
Publish
↓
No visible Google watermark
↓
Underlying AI provenance remains
This provides greater flexibility without completely removing Google’s ability to identify the content.
The Move Could Benefit Commercial AI Use
Companies increasingly use generative AI for marketing, advertising, product concepts, social-media campaigns and entertainment.
Visible watermarks can sometimes make AI-generated assets look less polished or interfere with branding.
The new option could therefore encourage broader commercial use of Google’s generative tools.
Commercial AI Content
AI generation
↓
Marketing asset
↓
Editing
↓
Brand integration
↓
Final publication
↓
No visible Google watermark
The underlying provenance systems can remain intact while the visible presentation becomes more flexible.
But Removing Visible Watermarks Raises Transparency Concerns
The change also raises questions about whether viewers will find it harder to identify AI-generated content.
A visible watermark provides an immediate signal that does not require a specialised detection tool.
If the mark is removed, someone viewing the content may not know that it was created using AI.
Transparency Challenge
Visible watermark removed
↓
Viewer sees no obvious AI label
↓
Content may appear human-created
↓
Need for invisible detection
↓
Need for verification tools
This is one reason Google is keeping SynthID and C2PA in place.
Invisible Watermarks Become More Important
As visible labels become optional, invisible identification systems could play a larger role in AI transparency.
These systems allow platforms and verification tools to inspect content without requiring a permanent visible label.
Visible vs Invisible
| Feature | Visible Watermark | Invisible Watermark |
|---|---|---|
| Visible to users | Yes | No |
| Affects appearance | Yes | Generally no |
| Immediate AI signal | Yes | No |
| Verification capability | Limited | Yes |
| Google option | Can be disabled | SynthID remains |
| Main purpose | Visible disclosure | Provenance and detection |
Google’s approach effectively prioritises invisible identification while giving users control over visible branding.
C2PA Could Provide Another Verification Layer
C2PA metadata provides information about the provenance of digital content.
It can help establish details about how content was created or modified when the metadata remains intact.
Google’s decision to retain C2PA information means removing the visible watermark does not necessarily remove all provenance information from the generated file.
Provenance Stack
AI generation
↓
C2PA metadata
+
SynthID
↓
Content provenance
↓
Verification tools
The approach is consistent with a broader industry movement toward machine-readable AI-content provenance.
Google Is Also Supporting Developer Verification
Google has announced an open-source library called Credentio that allows developers to implement local validation for content provenance.
The move could make it easier for third-party applications to verify C2PA-related information without depending entirely on Google’s own platforms.
Developer Ecosystem
Google AI generation
↓
Provenance information
↓
C2PA
↓
Credentio
↓
Third-party applications
↓
Independent verification
This could help extend AI-content verification beyond Google’s own services.
The Change Comes Amid a Broader AI Watermarking Debate
Technology companies are increasingly debating how AI-generated content should be labeled.
The industry has to balance two competing objectives:
- Making AI-generated content identifiable
- Allowing creators to use AI-generated material without intrusive labels
Google’s latest policy represents one approach to that problem.
AI Transparency Debate
Strong visible labeling
↓
High transparency
BUT
↓
Potentially less creative flexibility
OR
↓
Invisible identification
↓
Greater creative flexibility
BUT
↓
Less immediate transparency
Google is choosing to make the visible layer optional while preserving hidden identification mechanisms.
Other AI Companies Are Taking Different Approaches
Google’s decision comes as other AI companies are strengthening their own content-provenance systems.
Anthropic, for example, has announced invisible watermarking and metadata measures for content generated by its Claude models.
The different approaches reflect the rapidly changing regulatory and technological environment surrounding AI-generated content.
Industry Direction
↓
Optional visible watermark
+
SynthID
+
C2PA
Anthropic
↓
Invisible watermarking
+
C2PA
↓
Industry-wide focus on machine-readable provenance
The broader trend suggests that invisible or machine-readable identification could become increasingly important.
Regulation Is Also Driving Change
AI transparency regulations are becoming more significant in major markets.
The European Union’s AI Act includes transparency requirements for AI-generated and manipulated content.
These rules are increasing pressure on AI companies to develop reliable ways of identifying synthetic media.
Regulation and AI Provenance
AI-generated content
↓
Transparency requirements
↓
Machine-readable identification
↓
Watermarking
+
Metadata
+
Provenance standards
↓
Greater accountability
Google’s retention of SynthID and C2PA allows it to maintain an identification layer while changing the visible presentation of its generated content.
Some Regions May Not Get the Option
The visible-watermark removal feature may not be available in every market.
Where local laws require visible labeling of AI-generated content, Google may need to retain the visible watermark.
This means users in different countries could have different controls.
Regional Availability
Google introduces feature
↓
Local regulations checked
↓
If visible labeling is legally required
↓
Watermark remains
OR
↓
If permitted
↓
User can disable visible watermark
Regulatory requirements will therefore influence how widely the feature can be used.
AI Detection Will Become More Important
As visible watermarks become optional, AI detection and provenance tools may become increasingly important.
Users, platforms and publishers may need to rely more heavily on machine-readable signals rather than simply looking for a visible mark.
Future AI Verification
AI-generated content
↓
No visible watermark
↓
Detection system
↓
SynthID verification
+
C2PA verification
↓
AI origin identified
This could eventually become a normal part of digital-content infrastructure.
There Are Limits to Provenance Systems
Invisible watermarks and metadata are useful, but they are not necessarily perfect.
Metadata can sometimes be removed or lost when files are processed by other applications.
Watermarking technologies can also face attempts to manipulate or attack the embedded signals.
This means provenance systems are likely to remain one layer of a broader content-authentication framework rather than a universal solution.
What It Means for AI Creators
For creators, the biggest immediate benefit is greater control over how generated media looks.
They can potentially produce AI-generated assets without a prominent visible Google mark while retaining access to underlying provenance technology.
Potential benefits include:
- Cleaner visual output
- More professional-looking assets
- Easier brand integration
- Greater flexibility in editing
- Better control over published media
- Continued AI-origin verification
The feature could therefore make Google’s AI-generation tools more attractive to professional users.
What It Means for Businesses
Businesses using Google’s AI tools could gain greater flexibility when incorporating generated media into advertising and marketing campaigns.
Companies may be able to use AI-generated visuals without having the visible watermark interfere with branding.
However, businesses will still need to consider disclosure requirements in markets or industries where AI labeling is required.
Business AI Workflow
Generate
↓
Edit
↓
Brand
↓
Check disclosure requirements
↓
Publish
↓
Retain provenance
The regulatory environment will determine how businesses should use the new capability.
What It Means for Viewers
For viewers, the change could make AI-generated content harder to identify at first glance.
A visible watermark is a simple signal that requires no technical knowledge.
Once it disappears, viewers may need to rely on platform labels, provenance information or dedicated verification tools.
Viewer Experience
Before
↓
Visible AI watermark
↓
Immediate recognition
After
↓
No visible watermark
↓
Potentially harder to identify
↓
Use provenance and detection tools
This makes reliable verification infrastructure more important.
What It Means for Google
The change gives Google an opportunity to make its AI tools more useful to creators while maintaining its broader provenance strategy.
It also allows the company to respond to concerns that visible watermarks can interfere with creative output.
Google’s Strategy
Creative flexibility
+
Invisible provenance
+
C2PA metadata
+
Developer verification
+
Regulatory compliance
↓
More flexible AI ecosystem
The policy attempts to address both usability and transparency.
Key Features at a Glance
| Feature | Status |
|---|---|
| Remove visible watermark | Available/rolling out |
| Gemini | Supported |
| Flow | Supported |
| Google Search | Coming later |
| Nano Banana | Covered |
| Omni | Covered |
| Lyria | Covered |
| SynthID | Remains active |
| C2PA metadata | Remains |
| Credentio | Open-source developer validation library |
| Regions with mandatory visible labels | May retain watermark |
Infographic: Google’s New AI Watermark Approach
GOOGLE AI GENERATION
↓
IMAGE + VIDEO + MUSIC
↓
VISIBLE WATERMARK
↓
USER CAN TURN IT OFF
↓
CLEANER FINAL OUTPUT
BUT
↓
SYNTHID
↓
STILL PRESENT
+
C2PA METADATA
↓
STILL PRESENT
↓
AI ORIGIN CAN STILL BE VERIFIED
What Users Should Know
The key point for users is that removing Google’s visible watermark does not make the content “unmarked” in every sense.
The visible label can disappear, but Google’s invisible provenance technologies remain.
This distinction is important for creators who want a clean output and for viewers who need to understand whether media was AI-generated.
What Investors Should Watch
For Google, the policy change could increase adoption of its AI-generation products among professional creators and businesses.
Investors and industry observers should watch:
- Gemini media-generation adoption
- Flow usage
- Professional AI-content workflows
- Enterprise AI adoption
- AI advertising use
- SynthID adoption
- C2PA ecosystem growth
- AI transparency regulations
- Competitor watermarking policies
Google’s AI Media Strategy
AI models
↓
Creative tools
↓
More user control
↓
Professional adoption
↓
Enterprise use
↓
Larger AI ecosystem
The success of the change will depend partly on whether increased flexibility translates into greater usage.
The Bigger Picture
Google’s decision reflects a broader shift in how the technology industry approaches AI-generated content. Visible watermarks are easy for people to understand, but they can interfere with creative and commercial workflows. Invisible watermarks and provenance metadata offer a different model in which AI origin can be verified without permanently changing the appearance of the content.
The challenge is ensuring that these hidden systems remain reliable and accessible to users and third-party platforms. As AI-generated images, videos and music become increasingly realistic, the ability to establish where content came from will become more important. Google’s combination of optional visible labeling, SynthID and C2PA represents an attempt to balance creative freedom with the need for transparency.
Looking Ahead
Google’s decision to let users remove visible watermarks from AI-generated images, videos and songs gives creators greater control over the final appearance of their work. The option is being introduced across Gemini and Flow for supported models such as Nano Banana, Omni and Lyria, with Google Search expected to receive the functionality later. The change could make AI-generated media more practical for advertising, design, entertainment and other professional applications where a prominent watermark may interfere with the finished product.
At the same time, Google is not abandoning AI-content provenance. SynthID and C2PA metadata will remain in generated media, allowing supported systems to verify that content originated from Google’s AI tools. This creates a two-layer approach in which visible labeling becomes optional while invisible identification remains. As governments introduce stricter AI-transparency requirements and AI-generated content becomes harder to distinguish from human-created work, the success of Google’s strategy will depend on how reliable and widely accessible its hidden verification systems become.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
