Remove AI Watermark: Pixels vs Metadata

Not with a metadata tool, and ScrubPhoto does not try. An AI watermark is part of the image itself, either a visible badge drawn into the pixels (like Samsung's Galaxy AI mark or Meta's Imagined with AI) or an invisible pattern like Google's SynthID. The AI tags in the file's metadata, C2PA Content Credentials and IPTC fields, are a separate layer, and ScrubPhoto shows and removes those for free without uploading anything.

Check and scrub your file here

It runs in your browser. Nothing is uploaded, and the result is lossless.

Choose photos or videos

Drop files here or paste an image. Photo on your phone? Open scrubphoto.com there.

No upload. Works offline once the page loads.
Lossless. Every pixel kept, byte for byte.
Free. No sign-up, no limits.

No photo handy?

AI watermark vs AI metadata: the difference

A watermark lives in the pixels of an image or the frames of a video, either as a visible badge or as an invisible pattern that takes special software to detect. Taking it away means changing the picture itself.

Metadata sits next to the image data, in its own block of bytes. A C2PA manifest, also called Content Credentials, can name the app that made a file and say AI was used. IPTC fields can carry a digital source type like "created by AI". A tool can lift these blocks out without touching a single pixel.

Many AI makers use both layers on purpose. OpenAI's help page on provenance signals says platforms, editing tools and file conversions may remove metadata, while a watermark is embedded in the content itself.

Platform labels mostly come from the metadata layer. Meta says it labels a post when it finds industry standard AI indicators in the file, meaning C2PA and IPTC. That is what puts the AI info label (Instagram has also shown it as AI content, and first called it Made with AI) on a post. TikTok reads Content Credentials too.

Which apps add a visible AI watermark

The makers describe these visible marks on their own pages.

Samsung Galaxy AI. Samsung's support page on generative photo editing says images generated by AI include an AI watermark in the bottom-left corner. Its Galaxy S24 launch notes say a watermark goes into the metadata as well. Our Samsung watermark guide covers both.
Meta AI. Instagram's help page on labeling AI content says some content made with Meta AI tools has a visible watermark that says Imagined with AI.
Gemini Apps. Google's watermark settings page says visible watermarks appear on newly generated images and as a persistent overlay on generated video. It describes a setting for that visible mark and says the setting never affects SynthID or Content Credentials.

Which apps add an invisible AI watermark

These invisible marks are confirmed on official pages.

Maker	Invisible watermark	Metadata it also adds
Google Gemini Apps	SynthID in all media created or edited	Content Credentials (C2PA)
OpenAI (ChatGPT, Codex, API images)	SynthID	Content Credentials
Meta AI	Meta's own watermark	IPTC
TikTok AI tools and C2PA uploads	TikTok's own, readable only by TikTok	C2PA on AI content made on TikTok
Apple Intelligence (update later in 2026)	SynthID on adjusted photos	Not stated

Google DeepMind's SynthID page says the watermark is embedded directly into images, audio, text or video and cannot be perceived by people. TikTok announced its watermark in November 2025, and Apple announced SynthID for photos adjusted with Apple Intelligence in June 2026.

Remove AI watermark from image: what ScrubPhoto removes

ScrubPhoto works on the metadata layer only. The checker shows what a file carries, from C2PA and the app that signed it to the IPTC source type, XMP, EXIF and PNG prompt text. The home tool then removes all of it in your browser. Pick photos, read the verdict for Instagram, TikTok and Facebook, tap Scrub, then Share or Download.

For JPEG, PNG and WebP the result shows "Pixels identical", because the image data is never decoded. That cuts both ways: your photo keeps every pixel, and so does any badge or invisible watermark inside it. If a file came from one of the tools above, assume the watermark is still there after a scrub, and remember that no tool can promise a post stays unlabeled.

Remove AI watermark from video

Video works the same way. A visible overlay, like the one Gemini Apps can place on generated clips, sits in the frames, and SynthID can be embedded in video as well.

The video metadata tool blanks the metadata boxes in MP4, MOV and M4V files in place, including C2PA, XMP and Apple's GPS location. It re-encodes nothing, so the video and audio data are untouched.

Video also carries a duty that no scrub changes. Meta requires people to disclose photorealistic video and realistic audio that was digitally created or altered, and may apply penalties if they do not. TikTok requires labels on realistic AI content, and YouTube requires creators to disclose when AI meaningfully alters or generates photorealistic content.

What happens if you crop or edit out a visible AI badge

We will not explain how to do it, but here is what the makers and platforms say.

The badge is a disclosure, and Samsung says its mark is overlaid when you save to indicate the image was generated by AI. Taking it off hides that fact from viewers without changing what the image is.

The hidden layers usually stay. OpenAI says SynthID is designed to withstand common changes, including light edits and cropping. Instagram's help page says Meta AI content that is downloaded and uploaded again still gets the AI content label. Meta, TikTok and YouTube all say their own systems can detect AI and add a label.

The rules still apply. Canva's AI Product Terms, effective June 26 2026, prohibit using its AI tools to mislead anyone that content is human-made, or to remove provenance or metadata tags, including C2PA. Google's generative AI use policy bans claiming generated content was made solely by a human in order to deceive. TikTok says it may remove unlabeled AI content that breaks its Community Guidelines.

If you want a picture with no AI badge, post an original you made without AI. If you post the AI version, label it.

When removing AI metadata is legitimate

Privacy comes first, because a phone photo can carry GPS location, the camera model and serial numbers, and posting it can hand all of that to strangers.

Wrong labels come second. Instagram's help page says photo tools that use AI to change the size or color of an image may add a signal that the content was modified with AI. A real photo with a quick AI color fix is still a real photo.

Making an AI image pass as real is not on that list. ScrubPhoto only removes data, and it never writes fake camera details into an AI image. The limits page lists everything it cannot do, and the Instagram page explains how labels work there.

Questions about AI watermarks

Can you remove an AI watermark for free?

Not with ScrubPhoto or any other metadata tool, because the watermark is part of the image. What ScrubPhoto removes for free is the AI metadata, like C2PA Content Credentials and IPTC fields, in your browser with no upload and no account.

Does removing metadata remove an AI watermark?

No, because they are separate layers. A scrubbed JPEG compares pixel-identical to the original, so a visible badge or an invisible watermark like SynthID is exactly where it was. Our SynthID guide explains the invisible side.

How do I remove the AI watermark on a Samsung photo?

ScrubPhoto leaves it alone, because the mark is how Samsung discloses that an image was generated by AI. Samsung's edit steps end with saving a copy, so your original should still be in Gallery, and you can post that one instead.

Can you remove an AI watermark from a video?

Not the watermark itself, because a visible overlay or an invisible watermark lives in the frames. ScrubPhoto only blanks the metadata boxes in MP4, MOV and M4V files, clearing C2PA and location data while the picture stays as it was.

Is it illegal to remove an AI watermark?

This page is not legal advice, and laws differ by country. What clearly applies are the terms of the tools and platforms you use: Canva forbids removing provenance tags from AI content, and Meta, TikTok and YouTube can act on missing disclosures.

How can I tell if an image has an AI watermark?

A visible badge you can see. For invisible ones, Google says Gemini, Search and Chrome can check for SynthID, and OpenAI's help page points to openai.com/verify for its own signals. The ScrubPhoto checker covers the metadata layer without changing the file.

Sources

Official pages this page was checked against in October 2026: