The Gemini watermark is now optional. On 14 August 2026, Google announced that users will be able to remove the visible watermark from its AI generations — the small badge stamped on images, videos and songs produced by its models. Josh Woodward, Google’s VP for Gemini, confirmed the change, and the toggle is rolling out over the coming days.
It is a bigger shift than it sounds. Since the visible mark first appeared, every picture from Gemini’s AI image generation has carried an unmissable signal that a machine made it. Turning the Gemini watermark into a setting hands that decision to the user — and leaves the job of identifying AI content to layers most people never see.
This guide covers exactly what Google announced, how the new Gemini watermark toggle works, what SynthID and C2PA metadata still record, who gets the option and where, and what the change means for businesses, creators and anyone who needs to trust what they see online.
Table of contents
- What Google Announced About the Gemini Watermark
- How the Gemini Watermark Toggle Works
- What Stays Behind: SynthID and C2PA Metadata
- Why Google Dropped the Visible Watermark Requirement
- Who Gets the Toggle: Regional Rules and Subscriptions
- Models and Surfaces Covered at Launch
- The Risks: Provenance, Deepfakes and Trust
- What the Gemini Watermark Change Means for Businesses and Creators
- How to Verify AI Content Without the Visible Mark
- FAQ: Gemini Watermark Removal
- References
What Google Announced About the Gemini Watermark
The announcement came on Friday 14 August 2026, delivered by Josh Woodward, Google’s VP for Gemini, and first reported widely by TechCrunch. The visible Gemini watermark — until now a permanent fixture on AI-generated media — becomes a user-controlled setting.
The timing is no accident. Rival generators from other labs ship clean output by default, and Google’s badge had become a competitive disadvantage among exactly the professional users its paid plans court. By making the Gemini watermark optional while keeping its invisible tracking intact, Google is trying to hold two positions at once: friendlier to creators, and still accountable to regulators who expect AI content to be identifiable.
The scope: images, video and music
The toggle applies across the models behind Gemini’s media generation: the Nano Banana image models, the Omni video model and the Lyria music family. In practice that means the visible mark can be switched off on AI images, AI video clips and AI-generated songs alike, rather than on pictures alone.
Where the setting lives
The control appears under Settings, in a new Media Watermark section, inside the Gemini app and in Flow, Google’s AI video editor. Support in Google Search’s creation features is promised to follow soon. Once the rollout reaches an account, the visible mark on new generations follows whatever the user has chosen.
What Google is not removing
The company was explicit that Gemini watermark removal only affects the visible layer. The invisible SynthID watermark and the C2PA provenance metadata remain embedded in every generation, with no user-facing switch of any kind. Google’s position is that transparency moves from the surface of the file to its underlying signal.
How the Gemini Watermark Toggle Works
The mechanics are deliberately simple. Users open Settings, choose Media Watermark, and turn the visible mark on or off. The choice then applies to new AI generations made in that product. Nothing needs to be edited, exported or re-uploaded — the Gemini watermark simply is not drawn onto the output in the first place.
A default that still favours disclosure
The visible mark remains on by default. Google is not stripping badges from anything retroactively; it is letting individual users opt out for their own new creations. Anyone who never opens the setting will keep producing marked media exactly as before, which matters for the millions of casual users who never touch preferences.
Rollout timing
Woodward said the setting is rolling out “in the coming days” rather than arriving everywhere at once. Accounts will see the Media Watermark option appear progressively, with the Gemini app and Flow first in line and Search following later. Staggered rollouts are standard for Gemini features, and this one also has regional strings attached, covered below.
Why a toggle and not a removal tool
Positioning matters here. Google has not shipped a watermark eraser; it has stopped forcing a label. The distinction sounds subtle, but it keeps Google away from the business of scrubbing marks off existing files — a practice that caused the company real embarrassment in 2025, when people discovered its own models could be prompted to strip watermarks from other sources’ images.
What Stays Behind: SynthID and C2PA Metadata
The heart of Google’s argument is that the visible badge was always the shallowest of three transparency layers. A generation from Gemini carries the visible mark, an invisible SynthID watermark woven into the pixels or audio itself, and C2PA metadata — a signed record of the content’s origin following the standard from the Coalition for Content Provenance and Authenticity. The new toggle removes exactly one of those three layers; the other two stay on every file, always.
| Layer | What it is | Who can see it | Can users turn it off? |
|---|---|---|---|
| Visible Gemini watermark | Badge drawn on the corner of the media | Anyone looking at the file | Yes — the new toggle |
| SynthID | Invisible watermark embedded in pixels or audio | Detection tools only | No |
| C2PA metadata | Signed provenance record attached to the file | Software that reads content credentials | No |
Put as arithmetic: two of the three provenance layers — 67 percent of the stack — survive even when the visible mark is switched off.
SynthID at scale
SynthID is not a small experiment. Developed by Google DeepMind, it launched in 2023, and by Google I/O in May 2025 the company said more than 10 billion pieces of content already carried the invisible mark. It survives common edits such as cropping, compression and filtering far better than any corner badge, which was trivially cropped away.
C2PA and the newly open-sourced Credentio
Alongside the announcement, Google open-sourced Credentio, a C2PA library that lets developers build local content-credential validation into their own applications. That matters because C2PA metadata is only useful if software actually reads it — and Google clearly wants checking provenance to become as routine as checking a certificate on a website.
Why Google Dropped the Visible Watermark Requirement
Google did not frame this as a retreat from transparency, and the reasoning deserves a fair hearing before the risks are weighed.
Creators asked, competitors listened
Professional users have complained about the badge since day one. A visible mark in the corner of client work, a music video or a product mockup reads as a vendor logo, and creators either cropped it, paid for third-party removal tools, or switched to competing generators that never stamped output at all. The Gemini watermark toggle removes a reason to leave Google’s ecosystem.
The mark was already security theatre
Any determined user could crop, inpaint or regenerate away a corner badge in seconds — a whole cottage industry of removal sites proved it. In March 2025, reports showed people using Google’s own Gemini 2.0 Flash model to strip watermarks from images, including copyrighted photography. A signal that only honest users preserve mostly labels the honest. The invisible layers, which do not depend on user cooperation, were always the load-bearing part.
Regulation reads the metadata, not the badge
Rules such as the EU AI Act’s transparency obligations — which we analysed when the EU moved to mandate labels on authentic-looking AI content — are largely satisfied by machine-readable provenance and detection capability, not by a specific pixel badge. Keeping SynthID and C2PA on every file lets Google argue compliance while giving users the clean output they want.
A precedent other AI labs will study
Google is the first major lab to walk back a visible mark it once presented as a trust feature, and the rest of the industry will be watching how the Gemini watermark decision lands. If it passes without regulatory pushback, expect competitors that still stamp their output to follow within months. If lawmakers respond badly, the visible badge could yet return as a legal requirement rather than a design choice — which is exactly why Google kept the machine-readable layers untouched.
Who Gets the Toggle: Regional Rules and Subscriptions
Gemini watermark removal is not uniform worldwide. In most countries, the Media Watermark setting is available to users on free and paid plans alike. In India, South Korea and Vietnam, however, reports indicate the toggle is reserved for subscribers to the top-tier Google AI Ultra plan — everyone else in those three markets keeps the visible mark automatically.
| Where you are | Plan | Visible watermark |
|---|---|---|
| Most countries | Free or any paid plan | Optional — toggle in Settings |
| India, South Korea, Vietnam | Google AI Ultra | Optional — toggle in Settings |
| India, South Korea, Vietnam | Free and other plans | Stays on — no toggle offered |
Why those three markets differ
Google has not published its reasoning, but all three countries have been moving toward strict rules on labelling synthetic media, including India’s proposed requirements for prominent AI-content labels. Gating the Gemini watermark toggle behind a paid, identity-linked subscription in those markets looks like a compliance hedge rather than a pricing decision.
What it means for global teams
If your organisation produces AI media across regions, the same prompt can now yield differently labelled output in different offices. A marketing team in London can publish unmarked visuals while colleagues in Mumbai cannot — something worth writing into brand guidelines before inconsistent assets start shipping.
Models and Surfaces Covered at Launch
The change spans Google’s three generation model families and, at launch, two of its three announced surfaces. The Gemini app and Flow get the setting in the current rollout; Search’s creative features are next. That is 67 percent of the announced surfaces live on day one, rising to 100 percent when Search support ships.
| Model | What it generates | Where the toggle applies | Status |
|---|---|---|---|
| Nano Banana | Images | Gemini app | Rolling out now |
| Omni | Video | Gemini app and Flow | Rolling out now |
| Lyria | Music and songs | Gemini app and Flow | Rolling out now |
| All of the above | — | Google Search creation features | Coming soon |
Older generations are unaffected
The setting governs new output only. Media generated before the rollout keeps whatever marks it shipped with, and Google is not offering any tool to reprocess old files. For archives and asset libraries, the badge you already have is the badge you keep.
The Risks: Provenance, Deepfakes and Trust
Critics were quick to note the obvious: the visible mark was the only signal ordinary viewers could check without tools. Removing it shifts the burden of verification from the person scrolling to the platform hosting.
The viewer no longer gets a free clue
Most people will never run a detector on a suspicious clip. For them, the corner badge was the entire transparency system. Once the Gemini watermark is off, a synthetic image is indistinguishable at a glance from a photograph — precisely the property that makes generated media useful to advertisers and dangerous to information ecosystems, from fake product photos to political imagery.
Detection tools become critical infrastructure
The invisible layers only help if platforms actually check them. SynthID detection exists in Gemini and Search, and C2PA support is growing across creative software, but coverage is nowhere near universal. Social networks are still building likeness and provenance systems — TikTok’s approach, which we covered in our piece on AI likeness detection, shows how early this tooling remains. Until checking is automatic everywhere media is consumed, unmarked AI content will routinely pass as authentic.
A new asymmetry for fraudsters
For cybersecurity teams, the practical concern is speed: scam assets that once needed a manual cropping step now come out of a mainstream tool clean by default, at no cost, on a free plan in most countries. Nothing about the underlying fraud playbook is new, but one small point of friction — and one easy tell for training staff to spot fakes — is gone.
What the Gemini Watermark Change Means for Businesses and Creators
For legitimate users, the change is straightforwardly good news, with a few new responsibilities attached.
Cleaner assets, clearer policies
Marketing teams get client-ready output without badge-cropping workflows or third-party removal tools of dubious provenance. The sensible response is a written policy: decide which classes of content your organisation will publish unmarked, which keep the visible label for audience trust, and who owns the toggle. Disclosure that used to be automatic is now a brand decision.
Your audience may still expect a label
Platforms and regulators increasingly require synthetic-media disclosure regardless of what the file looks like. An unmarked AI image is still AI — and still detectable through SynthID by anyone using Google’s checking tools. Brands that quietly drop labels their audience expects are gambling with trust that is expensive to rebuild, as every deepfake controversy has demonstrated.
Watch the verification side of your stack
If your business consumes media — user submissions, influencer content, supplier assets — the disappearance of visible marks means your intake process needs a detection step. Building C2PA reading into workflows via libraries like Credentio, and using Gemini or Search to query whether an image is AI-generated, moves from nice-to-have to necessary.
How to Verify AI Content Without the Visible Mark
The good news: checking is genuinely possible, today, without specialist skills.
Ask Google’s own tools
Woodward pointed out that users “can still use Gemini or Search to see if an image was AI-generated.” Uploading an image to Gemini and asking whether it is AI-made triggers a SynthID check against Google’s models. The same capability is surfacing in Search. It is not a universal detector — it speaks for Google’s generations, not everyone’s — but it covers the exact content this toggle affects.
Read the content credentials
C2PA metadata travels with the file, and a growing set of tools — from Adobe’s Content Credentials inspector to anything built on Credentio — will display it. If a file’s credentials say it originated from a generative model, no visible badge is needed to know its history. If the credentials have been stripped, that absence is itself a signal worth treating with suspicion.
Train people for the new default
Teams that were taught “look for the corner badge” need an updated briefing: assume nothing from a clean corner. Provenance now lives in tools, not pixels. Building that reflex — check before you trust, especially for high-stakes media — is the single cheapest defence any organisation can deploy this quarter.
FAQ: Gemini Watermark Removal
How do I remove the Gemini watermark from AI images?
Open the Gemini app or Flow, go to Settings, choose Media Watermark, and switch the visible mark off. New generations from Nano Banana, Omni and Lyria will then render without the badge. The setting is rolling out gradually from 14 August 2026, so it may take days to reach your account.
Does turning off the Gemini watermark remove SynthID too?
No. The toggle affects only the visible layer. The invisible SynthID watermark and the C2PA provenance metadata are embedded in every generation regardless of the setting, and there is no user-facing way to disable either. Google’s detection tools can still identify the content as AI-generated.
Is the Gemini watermark toggle free?
In most countries, yes — it is a settings option on free and paid plans alike. In India, South Korea and Vietnam, reports say the option is limited to Google AI Ultra subscribers, and content generated on other plans in those markets keeps the visible watermark automatically.
Can I tell whether an unmarked image came from Google’s AI?
Yes, in most cases. Upload it to Gemini or use Google Search’s checking features and ask whether it is AI-generated; a SynthID match confirms it came from Google’s models. You can also inspect the file’s C2PA content credentials with any compatible tool, including software built on Google’s newly open-sourced Credentio library.
Which products support the Gemini watermark setting today?
At launch the Gemini watermark toggle appears in the Gemini app and in Flow, Google’s AI video editor, covering the Nano Banana, Omni and Lyria model families. Google says the setting will reach its Search creation features soon. The rollout is gradual, so the Media Watermark option may take a few days to appear in your settings.
References
TechCrunch: Google will now allow users to remove visible watermark from its AI generations
Digital Trends: Google’s Gemini app adds a toggle to disable AI watermarks, with some exceptions
Android Authority: Google might soon let you skip the Gemini watermark on your Nano Banana images
Android Headlines: Google Gemini Finally Lets You Disable Visible AI Watermarks
Google: SynthID Detector — a new portal to help identify AI-generated content
Coalition for Content Provenance and Authenticity (C2PA)
TechCrunch: People are using Google’s new AI model to remove watermarks from images
