The Gemini Watermark: Two Marks, Only One Of Them Removable

Every image Gemini creates or edits carries two separate marks. One is the visible sparkle logo sitting in a corner of the picture. The other is SynthID, an invisible signal woven into the pixels.

Almost every tool sold as a Gemini watermark remover targets the first mark. Google's own documentation says the second one stays regardless, and that it is built to survive rescaling, recoloring and compression.

Search demand for this question is heaviest outside the United States, and a large share of it arrives in Hindi and other non-English queries. The mechanics are the same everywhere, with two regional exceptions covered below.

This guide explains what each mark is, which Gemini surfaces apply which, whether paying removes the visible one, and how to check an image with Google's own verification tools. It does not walk through removal steps.

Gemini Puts Two Different Marks On Every Image

Gemini applies a visible watermark and an invisible watermark, and they are unrelated technologies. The visible one is a logo drawn on top of the picture. The invisible one is SynthID, a statistical signal embedded inside the image data by Google DeepMind.

The confusion behind most searches comes from treating these as one thing. Removing the corner logo does nothing to the SynthID signal, because the signal is not stored in that corner. It is spread across the whole image.

There is a third layer on Gemini Apps media as well. Google attaches C2PA Content Credentials, the signed provenance metadata standard, to media created in Gemini Apps.

So a single Gemini image can carry a logo you can see, a signal you cannot see, and a metadata record describing how the file was made. Each behaves differently when the file is edited.

• Visible sparkle: a logo overlay on the image, video frame, or music player.
• SynthID: an invisible signal inside the pixels, applied to all Gemini Apps media.
• Content Credentials: signed C2PA metadata attached alongside the file.
• Not interchangeable: stripping one leaves the other two untouched.

The Visible Sparkle Mark And Where It Appears

The visible mark is the Gemini sparkle logo, placed on media generated in the Gemini apps. Google describes how it lands differently by media type. On images it appears on newly generated pictures.

On video it appears as a persistent overlay across the generated clip, not a single frame. On music it appears on the video or visual player created with the track rather than on the audio itself.

This mark is a graphic composited onto the output. It occupies specific pixels, so cropping it away or painting over it produces an image with no logo. That is the entire technical basis of the remover tool category.

Coverage is not uniform across Google's products. The Gemini Apps watermark setting applies to the Gemini web and mobile apps only, not to AI features built into other Google products, which follow their own rules.

• Images: visible mark on newly generated images.
• Video: a persistent overlay across the generated clip.
• Music: the mark appears on the visual player, not the audio file.
• Scope: the Gemini Apps setting does not govern other Google products.

Why SynthID Survives Cropping, Resizing And Screenshots

SynthID survives ordinary edits because it is not stored in one place. Google DeepMind embeds the signal across the image so that no single crop, filter, or resize removes it.

Google states the SynthID image watermark is designed to stand up to modifications like cropping, adding filters, changing frame rates, and lossy compression. Its Gemini support page says the mark will usually still exist after an image, video or audio file is rescaled, recolored or compressed.

Screenshots are the case people ask about most. Google's advertising documentation describes SynthID as resistant to manipulations including screenshots, filters and compression, because a screenshot reproduces the same pixel relationships the signal lives in.

The audio version works on the same principle. Google says the audio watermark is inaudible and is not altered by common changes such as added noise, MP3 compression, or speed changes.

None of this makes SynthID unbreakable. Google describes it as durable against common modifications, not as tamper-proof, and heavy enough destruction of the image will take the signal with the quality.

• Distributed, not local: the signal spans the image rather than a corner.
• Designed to survive: cropping, filters, frame-rate changes, lossy compression.
• Screenshots: Google lists screenshots among the manipulations SynthID resists.
• Audio: unaffected by added noise, MP3 compression, or speed changes.
• Not absolute: Google claims durability against common edits, not immunity.

What Gemini Watermark Remover Tools Actually Remove

Tools marketed as Gemini watermark removers strip the visible sparkle logo. They crop the region, or they inpaint over it by generating replacement pixels that match the surrounding image.

Both approaches operate on the visible layer only. Neither reads nor targets the SynthID signal, because that signal is not a graphic element these tools can locate and paint over.

C2PA metadata is a different story again. Provenance metadata is strippable by design limits of the standard, so re-saving or converting a file can drop it, as our C2PA content provenance guide explains in detail.

That produces a specific and common failure. Someone crops the sparkle, re-saves the file, believes the image is now unmarked, and then Google's own checker still identifies it as created or edited by Google AI. The visible cue is gone; the machine-readable one is not.

There is also a quality cost that rarely gets mentioned. Inpainting invents pixels where the logo was, so fine detail, text, and straight edges near the mark often come back wrong.

• Removed: the visible sparkle logo, by cropping or inpainting.
• Usually dropped: C2PA metadata, on re-save or format conversion.
• Not removed: the SynthID signal inside the image data.
• Side effect: inpainting degrades detail around the former logo area.

Does Paying Remove The Gemini Watermark?

Paying has removed the visible mark for some users, and as of August 2026 Google also offers a setting for it. Neither route removes SynthID.

When Nano Banana Pro launched, Google said it would keep the visible Gemini sparkle on images from free and Google AI Pro tier users, and remove it for Google AI Ultra subscribers and inside the AI Studio developer tool.

Gemini Apps now carries a Media Watermark setting, found under Settings, with an on or off switch. Google's support page states plainly that this setting only controls the visible watermark, and that it does not affect SynthID watermarks or Content Credentials.

Two exceptions are worth knowing before you go looking for the toggle. Google says users in India, South Korea and Vietnam can only access the setting with an AI Ultra subscription, and that people on work or school accounts cannot change it at all.

Because these tier and setting details have already changed more than once, confirm the current behavior on Google's own watermark settings page rather than trusting a summary, including this one.

• Nano Banana Pro launch policy: sparkle kept for free and Google AI Pro, removed for Google AI Ultra and AI Studio.
• Now: a Media Watermark on/off setting in Gemini Apps settings.
• Regional exception: India, South Korea and Vietnam require AI Ultra for the setting.
• Managed accounts: work and school accounts cannot change it.
• Unchanged either way: SynthID and Content Credentials remain.

How To Check A Gemini Image For SynthID

You can check an image in the Gemini app itself. Google's documented method is to sign in at gemini.google.com, upload a single file, and ask whether the image was created or edited by Google AI.

Google publishes limits on this check. It accepts one file at a time up to 100 MB, and allows roughly ten image checks, ten video checks and ten audio checks per rolling 24-hour window.

Google also runs a separate verification portal, SynthID Detector, which scans an upload and highlights the portions most likely to be watermarked. Access has been offered to journalists, media professionals and researchers through a waitlist.

Read the results carefully, because the two outcomes are not symmetrical. A detection means the file was created or edited by Google's AI models. No detection only means no Google watermark was found, and Google says the content could still have been made by another AI system.

The checker can also return an inconclusive answer. Google lists responses such as not enough details to watermark and likely too small an edit, both of which mean the check could not decide rather than that the file is clean.

For the wider detection landscape, including tools that guess from style rather than read a real signal, see our AI watermark detector guide and our AI image detectors compared roundup.

• Where: the Gemini app, or the SynthID Detector portal for approved users.
• Limits: one file, up to 100 MB, around ten checks per media type per day.
• A hit proves: Google AI created or edited the file.
• A miss proves: nothing about other AI tools.
• Inconclusive results exist: too little detail, or too small an edit.

Editing Your Own Photo In Gemini Also Marks It

The marking rule catches a case most people do not expect. Google states that all AI-generated media created or edited with Gemini Apps includes an invisible SynthID watermark.

Edited counts. If you upload a photograph you shot yourself and use Gemini to retouch it, the file that comes back carries the signal. The image is yours, and it is still marked.

This is the single most useful thing to understand before you go hunting for a remover. The question is not whether the picture is original work. It is whether a Google model touched the pixels.

For teams, that turns into a workflow decision rather than a removal problem. Decide which images may pass through Gemini at all, and keep an unedited master of anything that must stay unmarked.

Google also announced it is adding C2PA Content Credentials verification in the Gemini app, with Search and Chrome to follow, which makes provenance easier to check outside Google's own apps over time.

• Generated or edited: both paths get SynthID in Gemini Apps.
• Your own photo: retouching it in Gemini marks the returned file.
• Practical control: keep an untouched master of images that must stay unmarked.
• Direction of travel: provenance checks are expanding into Search and Chrome.

Where The Legal Line Sits

Removing a visible mark from your own generated image is a different question from misrepresenting where content came from, and Google's Generative AI Prohibited Use Policy bars misrepresenting the provenance of generated content by claiming it was created solely by a human, in order to deceive. Legal exposure depends on your jurisdiction and use, and we cover it properly in our guide on whether it is legal to remove an AI watermark.

For which vendors mark what, and how Google compares to other labs, our roster of AI models that watermark their output is the page we keep current. The mechanics behind marking in general are covered in our AI watermarking pillar.