Guides
Can You Remove an AI Watermark From an Image? The Facts
AI images can carry a visible logo, metadata labels (C2PA, IPTC) and invisible watermarks like SynthID. What removing each one does and does not do.
Partly, and it matters which "watermark" you mean. AI images can carry three different marks: a visible logo, metadata labels (C2PA Content Credentials or the IPTC digital source type) and an invisible watermark hidden in the pixels, such as Google's SynthID. Stripping metadata removes the labels but not the invisible watermark, and none of it changes what the image is. If the goal is to pass an AI image off as a real photo, don't: platforms detect and label it anyway, and in many places it can break the rules or the law.
The three kinds of AI marks
| Mark | Where it lives | What stripping metadata does | Who can read it |
|---|---|---|---|
| Visible logo or label | In the pixels you can see | Nothing | Anyone looking at the image |
| C2PA Content Credentials | A signed manifest attached to the file | Removes it from that copy | Verify tools, some platforms and search engines |
| IPTC digital source type | A metadata field (for example trainedAlgorithmicMedia) | Removes it | Platforms, search engines, metadata viewers |
| Invisible watermark (SynthID and similar) | Spread across the pixels | Nothing | The provider's detector |
C2PA is an open standard: a manifest records who made the file, with which tool, and what was done to it, and it's cryptographically signed. See what C2PA Content Credentials are for the details.
IPTC digital source type is a single field from the IPTC photo metadata standard. Values like trainedAlgorithmicMedia (generated by an AI model) or compositeWithTrainedAlgorithmicMedia (a real photo edited with generative AI, such as inpainting) tell software how the image was made.
SynthID is Google's invisible watermark. Google says it is designed to withstand cropping, filters and lossy compression without affecting image quality. OpenAI's help pages say images from ChatGPT and its API now carry both C2PA metadata and SynthID.
What removing metadata actually does
Removing metadata, whether with a metadata tool, a re-save in another app, an upload to a platform that strips files, or a screenshot, has these effects:
- The C2PA manifest is gone from that copy. The signed history no longer travels with the file.
- The IPTC digital source type is gone. Platforms that read it can't label the image from metadata alone.
- The invisible watermark stays. It is part of the pixels. Detectors such as Google's SynthID Detector, the check built into the Gemini app, and OpenAI's verification page look for it, not for metadata.
- The content is unchanged. An image with no metadata isn't more "real". It just has less information attached.
C2PA was designed with this in mind: the specification describes "soft bindings", such as watermarks and fingerprints, that let a removed manifest be found again in cloud storage. Meta has said publicly that it reads C2PA and IPTC signals to label AI images, and that it is working on invisible watermarks precisely because metadata can be stripped.
Legitimate reasons to remove metadata
There are honest reasons to strip metadata from an image, AI-generated or not:
- Privacy. Metadata can include software, device, dates, edit history and sometimes location or account-linked details. You may not want that in a file you send.
- Clean files for work. A designer delivering a final asset may strip everything except what the client needs.
- Your own photos. A photo from your phone, lightly edited with an AI tool, may carry labels you'd rather not share along with its location.
Removing metadata for privacy is fine. What crosses the line is presenting an AI image as a real photo where that matters: news, evidence, documents, identity, or anything that involves a real person who didn't consent.
To see what a file contains before you share it, open it in our photo metadata viewer, which runs in your browser without uploading. Our AI image checker shows whether the file carries Content Credentials or an AI label. To clear EXIF, GPS and other metadata, the EXIF remover re-encodes the image locally. Both are about privacy; neither touches watermarks in the pixels.
What platforms and laws expect
Rules differ by platform and country. A few examples from primary sources:
- YouTube requires creators to disclose realistic altered or synthetic content, and may add a label itself if a creator doesn't.
- Meta (Facebook, Instagram, Threads) labels images it detects as AI-made using C2PA and IPTC signals, asks people to disclose realistic AI video and audio, and says it may apply penalties when they don't.
- The EU AI Act (Article 50) requires providers of generative AI to mark output in a machine-readable way and requires those who publish deepfakes to disclose that the content is artificial.
- Italy added a criminal offense (Article 612-quater of the penal code, introduced by Law 132/2025) for spreading AI-generated or altered images of real people without consent, made to look real, when it causes unjust harm.
None of this stops you from making or editing AI images. It's about being honest when a fake could mislead.
If you want an image without an AI label
Usually the honest route is also the simplest:
- Use a provider setting, not a hack. Some providers let paying users turn off the visible logo. That is their choice to offer; the invisible watermark and Content Credentials usually remain.
- Disclose instead of hiding. A caption such as "AI-generated illustration" removes the problem the label was meant to solve.
- Use a real photo when you need a real photo. For IDs, evidence, listings and news, an AI image isn't a substitute.
- Strip personal metadata, not provenance, when privacy is the goal. Check what's in the file first, then decide.
If you're on the other side, trying to work out whether an image you received is AI, see how to check if an image is AI-generated.
Where sigil. fits
sigil. isn't an AI tool and doesn't add or remove watermarks. It's an encrypted vault for the photos and documents you'd rather keep on your phone and away from other people's eyes. It strips EXIF and location metadata when you import and export photos, and photos taken with its private camera never land in the Camera Roll. If metadata is a privacy concern for you, how to check photo metadata explains what's in your own photos.
Bottom line
Stripping metadata removes C2PA and IPTC labels from a copy of an image. It doesn't remove an invisible watermark like SynthID, and it doesn't make an AI image real. Remove metadata for privacy, disclose AI images where it matters, and don't disguise them to deceive anyone.
Sources
- C2PA Technical Specification and Explainer (c2pa.org)
- IPTC Digital Source Type vocabulary
- SynthID, Google DeepMind
- Verify Google AI-generated images and videos with SynthID (Gemini Help)
- C2PA in ChatGPT images (OpenAI Help Center)
- Labeling AI-generated images on Facebook, Instagram and Threads (Meta)
- Disclosing use of altered or synthetic content (YouTube Help)
- EU AI Act, Regulation (EU) 2024/1689, Article 50 (EUR-Lex)
- Law 23 September 2025, no. 132 (Normattiva)
Frequently asked questions
Does removing metadata remove an AI watermark?
It removes metadata labels such as a C2PA manifest or an IPTC digital source type. It does not remove an invisible watermark like SynthID, which lives in the pixels and is designed to survive cropping, filters and compression.
Does a screenshot remove the AI label from an image?
A screenshot drops the original file's metadata, so C2PA and IPTC labels are usually gone. An invisible pixel watermark is designed to survive that kind of copy, and the visible content is unchanged.
Is it illegal to remove an AI watermark?
It depends on where you are and what you do with the image. Passing off an AI image as real to deceive people can break platform rules and, in some places, the law; for example, the EU AI Act requires deepfakes to be disclosed and Italy has a specific offense for harmful AI fakes of real people.
Why would I remove metadata from an AI image I made?
Privacy is a legitimate reason: metadata can include account details, software and edit history. Removing it for privacy is different from hiding that an image is AI-generated when that matters to the people who see it.
Can I check whether an image still has an AI label?
Yes. Content Credentials can be inspected with the free Verify tool at contentcredentials.org, and our photo metadata viewer shows the EXIF fields in your browser without uploading the file.