Can you remove SynthID from an image?
Not with any guarantee. SynthID is a watermark written into the pixels, and Google says it is designed to survive cropping, filters and compression. PassReal does not claim to remove it and gives no method against it. What can be removed is the C2PA and IPTC data that the same Google tools write into the file. The PassReal guarantee covers public AI detectors, not SynthID.
Last updated: 2026-10-08
The honest answer
You cannot count on removing SynthID. It is a watermark that Google writes into the pixels of images made with its generative tools, and the company says it is built to hold up under common changes. Nobody outside Google has published a reliable, independently tested way to take it out, and anyone who sells one with a guarantee is promising more than the evidence allows.
PassReal does not offer a SynthID remover. This page will not describe a way to attack the watermark either. What it can do is separate the things that cannot be removed from the things that can, so you know what you are dealing with.
What Google says
Google describes SynthID as a tool that embeds a digital watermark directly into AI-generated content. For images and video, its page says the watermark is designed to stand up to modifications such as cropping, filters and lossy compression. Those are design goals stated by Google. They are not a test result that we have been able to repeat, and we have no result of our own that says otherwise.
The watermark is meant to be found by a detector later. Google offers checks in Gemini, Search and Chrome, and a separate SynthID Detector. The mechanism is described in more detail in how the SynthID watermark works, which we do not repeat here.
Why metadata tools do not help
Many people try a metadata cleaner first and then find the image still reads as AI. That is expected. Metadata is a set of blocks beside the picture: EXIF, XMP, IPTC and C2PA. A watermark in the pixels is not in any of those blocks. Clearing the blocks leaves every pixel as it was, so the watermark stays.
It also works the other way. A check that says an image is AI-made does not prove the file still has metadata, and a clean file does not prove the pixels are free of a watermark.
What can be removed next to it
The same Google tools that add a pixel watermark also write standard provenance data into the file. That data is separate, and it can be removed:
- A C2PA manifest, the signed record of which tool made the image.
- The IPTC digital source type field, which names the image as made by a generative model.
- XMP tags and PNG text chunks that list the software.
This is the layer platforms read when they decide whether to show an AI label. Meta says it uses the standard indicators in the file for this. You can clear it by hand with exiftool -all= image.jpg and then read the file again to confirm. The steps are in how to remove C2PA from an image.
PassReal runs the same removal in one upload and reports the counts. On a Gemini export, 174 markers went to 0.
What the PassReal guarantee covers
PassReal's other stages work on the pixels, but they are built for a different target. They add sensor-style grain and a lens colour fringe, smooth the high frequencies with a Fourier filter, and re-encode the picture as a phone JPEG with matching EXIF. They aim at the signals that public AI image detectors measure. They were not designed against SynthID, and we do not say they remove it.
The guarantee follows that line. If a public AI image detector flags an image you paid to process, email us within 14 days with the result and the payment is refunded in full. The text is in the terms. A SynthID check from Google is a different test, and it is outside that promise.
How to decide
1. Ask what you need. If it is to stop a platform from showing an AI label, the file data is the part to deal with. 2. If a client, a contest or a platform needs to know an image is AI-made, plan on the watermark being checkable. 3. If you need certainty about SynthID on a particular file, ask the person selling the fix to show a result on your own image, with Google's detector as the judge. 4. Keep the original. A processed copy cannot give you the first file back.
Disclosure stays with you
Platform rules and local law can require you to say when an image is AI-generated. Taking out a marker does not remove that duty, and the duty rests with whoever posts the image.
Questions
Can you remove SynthID from an image?
No tool can promise it. Google designs SynthID to survive common edits, and we have no independent proof that any method removes it. PassReal does not offer to.
Does deleting metadata remove SynthID?
No. Metadata sits beside the image data and SynthID is in the pixels. You can clear EXIF, XMP, IPTC and C2PA and the watermark is still there.
What can PassReal remove from a Gemini image?
The provenance data in the file. On a Gemini export, 174 C2PA and IPTC markers went to 0. That does not touch a pixel watermark.
Does the PassReal refund guarantee cover SynthID?
No. The guarantee covers public AI image detectors: if one flags a paid, processed image, email us within 14 days and the payment is refunded. A Google SynthID check is not part of it.
Should I remove a watermark meant to show an image is AI-made?
That choice is yours, and so is the duty to follow platform rules and local law on declaring AI use. Removing a marker does not cancel the duty to disclose.
Limits
Naturalization operates on the pixels and on the file, not on the content of the image. A subject that is implausible - six fingers, inconsistent reflections, text that does not read - stays implausible after processing, and a human reviewer will notice it. PassReal changes what a statistical classifier measures, not what a person sees.
Results vary by generator, by subject and by detector, and detectors are retrained. No pass rate is published on this page because none has been measured in a way that would still hold next month.
Three images, no card, no expiry.