Pixel watermark · not metadata

Can you remove SynthID from an image?

No. You cannot remove SynthID from image pixels with this tool. SynthID is a watermark Google embeds in the picture. The page strips metadata beside it, such as EXIF, GPS, IPTC, XMP, and C2PA, and it does not edit the image data.

Strip EXIF, GPS, and C2PA

Strip mode

AI only leaves camera EXIF and GPS. All metadata removes them.

1 free image on this browser. No account. $8.99 one-time for unlimited images and one ZIP.

The cleaned file shows up here.

What SynthID is

SynthID is Google’s watermark written into the content. DeepMind describes it for AI-generated images, and also for audio, text, and video. The image watermark is meant to stay imperceptible when a person looks at the picture. Gemini Apps Help separates it from Content Credentials: SynthID is the invisible watermark; Content Credentials are C2PA metadata about origin. A file can carry both. The metadata can leave. The watermark is not a tag this cleaner lists.

That split is the whole answer to “can you remove SynthID” on this site. Metadata sits in segments beside the picture: EXIF, GPS, IPTC, XMP, a C2PA manifest, a PNG text chunk. SynthID sits in the pixels. A tool that rewrites segments and then re-scans those segments can show an empty After report and still leave the watermark untouched. This page will not describe a way to take the watermark out, and it will not quote a success rate for anyone else’s method.

What the inspect report can and cannot see

Drop a JPEG, PNG, WebP, HEIC, AVIF, GIF, TIFF, or BMP on the tool. Press Inspect. The report is a list of metadata categories this scanner knows. A C2PA line means a manifest is in this container. An AI line means a known hint such as IPTC DigitalSourceType or a generator text chunk. A GPS line means a location was stored with the file. No line means this scan did not find that category.

SynthID never appears as one of those lines. The scanner does not decode a pixel watermark, and a quiet report is not a verdict that the picture is free of one. The same quiet report is also not a verdict that a model made the picture. Extraction of file fields and detection of a watermark are different jobs. This site only does the first.

After you clean, the After column is a second scan of the new bytes, not a copy of the Before list. When the generation lines you saw are gone, the metadata job is done. The pixels of the picture are the pixels you dropped. That is not a SynthID process. Do not read a clean After column as “SynthID removed”.

A download that carries both marks

Gemini and Imagen exports are the case people hit first. Newer Google files can include C2PA Content Credentials, and the same image can carry SynthID. The metadata page for that download is Google Gemini and Imagen image metadata. Strip the manifest there if the report lists it. Leave this page if the only goal was the watermark.

OpenAI documents the same split for supported ChatGPT image downloads: file metadata plus SynthID in the pixels. The write-up is ChatGPT image metadata. Cleaning the download removes the metadata the report lists. It does not remove the watermark, and it does not force a platform to drop a label on the next upload.

Other generators are not the same object. Adobe Firefly JPEG and PNG exports are a Content Credentials case, written up on Adobe Firefly image metadata. A Photoshop Generative Fill export is the same family of file metadata, on Photoshop Generative Fill metadata. Midjourney downloads often have no C2PA chip at all. None of those pages is a SynthID remover. If a file from one of them also happens to carry a pixel watermark, this cleaner still does not touch it.

A visible mark is a third object. Copilot can print “Made with AI” into the picture. That mark is pixels, like a watermark you can see. The C2PA part of the same file is metadata. The page that separates those two is Copilot Made with AI label. Samsung’s Gallery text “AI-generated content” is also visible pixels, not a tag this tool deletes. The page is Samsung AI-generated content label.

What a screenshot, a crop, and a re-save are

A screenshot is a new image. It often drops the old file’s C2PA manifest and other metadata, because the new file is a picture of the screen rather than the original container. It does not remove SynthID. The measured statement on this site is does a screenshot remove AI metadata: C2PA usually drops, the pixel watermark stays, and a platform can still label the new image. Use that page when the question is the screenshot. Use this page when the question is the watermark.

A crop, a filter, a resize, or a re-export from an editor is an edit of the pixels or a new container. This tool does not crop, filter, or re-encode to chase a watermark. This page does not test those edits as a removal method, and it does not recommend them as one. Google describes SynthID as robust to many common modifications. This site does not turn that sentence into a number, and it does not publish a procedure that tries to beat it.

Command-line metadata wipes are the same boundary. We measured exiftool -all= on C2PA files and wrote down what else it removes. That write-up is ExifTool remove all metadata. It is a metadata command. It is not a SynthID command. The browser comparison is the ExifTool alternative.

Platform labels are a fourth object

Instagram, Facebook, TikTok, LinkedIn, Pinterest, and X can show a label after you upload. Some of those labels start from a tag in the file. Some start from a disclosure you set in the app. Some start from the platform’s own systems. Stripping metadata prepares the next file. It does not edit a post that is already live, and it does not switch off a classifier.

If the label you see is Instagram’s AI info, the owner is remove Instagram AI info, including the reasons a label remains after the file is clean. Facebook is remove the Facebook AI label. A TikTok still is remove the TikTok AI label. A video file is not one of those jobs. A cover or poster saved as JPEG, PNG, or WebP is remove AI metadata from a video cover. The MP4 or MOV stays closed, on that page and on this one.

LinkedIn’s CR icon is a Content Credentials mark, not SynthID. The removal path for the next still is remove the LinkedIn CR icon. A real photo can pick up AI fields in Lightroom without being a generated image. That case is Lightroom Content Credentials. Canva Background Remover can leave a real photo with a file signal Instagram later labels. That case is Canva AI content label. None of these are watermark removal.

What to do with the file you have

  1. Drop the JPEG, PNG, WebP, HEIC, AVIF, GIF, TIFF, or BMP. RAW and video are not opened here. A PDF uses remove metadata from a PDF. Word and Office files use remove metadata from a Word document.
  2. Press Inspect. EXIF, GPS, IPTC, XMP, and C2PA show up when they are in the file. SynthID does not.
  3. If the job was those tags, clean and download the file whose After report matches. Remove EXIF is the GPS and camera pass. Remove C2PA is the Content Credentials pass. The general cleaner is the AI metadata remover.
  4. If the only goal was SynthID, stop. This site does not remove it, and this page will not pretend to.

The vocabulary for the metadata half is what AI metadata is, including the section on ai metadata tagging and extraction. Tagging writes the file fields. Extraction reads them. SynthID is in neither step. GPS and camera fields, when those are the actual risk, are remove EXIF. A PDF is a different container: remove metadata from a PDF.

What this page will not give you

It will not name a remover, a script, a model, or a sequence of edits that claims to strip SynthID. Publishing that would be a removal method, and this site does not have one it can stand behind. It will not invent a detection rate, a surviving-edit percentage, or a count of files tested for the watermark. The only counts on the shutter-count pages are camera counters read from sample files. They have nothing to do with SynthID.

It will not tell you the watermark is gone because the After report is empty, because a screenshot looks clean, or because a platform has not labeled the upload yet. A missing label is not a watermark test. It will not tell you to use this tool where a platform or a law requires disclosure. Stripping metadata is a file edit. Hiding that a picture was generated is a different intent, and it is out of scope.

If you came here from a search for a SynthID remover, the honest product is the AI metadata remover and the AI metadata cleaner. Use those when the report lists C2PA, IPTC, XMP, or EXIF. Close the tab when the report is already empty and the watermark was the only thing you wanted gone. Questions about how the tool behaves in the browser are on how to use Remove AI Meta.

You cannot remove SynthID from image pixels here. The tool only strips metadata that sits beside the picture.

To read generator records without treating them as a watermark test, use check if image is AI.

Questions

Can this site remove SynthID from an image?

No. The cleaner strips metadata. SynthID is a watermark in the pixels, and this tool does not edit pixels.

Does removing EXIF or C2PA remove SynthID?

No. EXIF, GPS, IPTC, XMP, and C2PA sit beside the picture. SynthID does not. Clearing those tags leaves the watermark.

A Gemini or ChatGPT download has both. What actually gets removed?

Metadata the inspect report lists, such as C2PA, when you run a clean. SynthID stays. This page does not describe a way to strip the watermark.