Can you remove SynthID? The honest answer.
SynthID is an invisible watermark inside the pixels, not a piece of metadata. Here is what that means, what metadata removal can and cannot do, and how to check an image yourself.
What SynthID is
SynthID is Google DeepMind's watermarking system. Google describes it as a tool that "embeds imperceptible watermarks" in AI-generated content, and says it now covers text, audio, video and images from its Gemini, Imagen, Lyria and Veo models, with over 10 billion pieces of content watermarked (Google, May 20, 2025). OpenAI also adds SynthID watermarks, alongside C2PA metadata, to images made with ChatGPT and its API (OpenAI Help Center).
The key word is embedded. Metadata is stored next to the image, in separate segments of the file. A SynthID watermark is a pattern spread through the pixel values themselves. There is no "SynthID box" in the file to delete.
Why metadata tools can't remove it
A metadata remover, including this one, works on the parts of a file that are not the picture. That is exactly why it can promise identical pixels, and it is also why it cannot touch anything that lives in the pixels. Any tool that claims to remove SynthID has to change the image itself, and Google designed the watermark to survive common edits such as cropping, filters and compression.
| C2PA / IPTC metadata | SynthID watermark | |
|---|---|---|
| Where it lives | Separate segments of the file | Inside the pixel (or audio) values |
| Visible? | No, but readable with any metadata viewer | No, and not readable without Google's detector |
| Removed by this tool? | Yes | No |
| Who adds it | OpenAI, Adobe and others, per the C2PA standard | Google models; OpenAI images |
How to check for SynthID
Google runs the SynthID Detector, a portal where you can upload an image, audio, video or text and get an assessment of whether a SynthID watermark is present. It can also highlight the parts most likely to carry the watermark (Google).
What this means for AI labels
The platforms that explain how they label content describe reading metadata: Meta reads C2PA and IPTC "AI generated" fields (Meta) and TikTok reads Content Credentials (TikTok). Removing that metadata clears those signals. A pixel watermark is a separate signal, and whether any platform checks it is up to that platform.
If an image is your own work and was only touched up by an AI tool, removing metadata is usually what you need. If an image was fully generated by a model that adds SynthID, the watermark stays with it, and in many places the right move is simply to disclose it.
Common misconceptions
"Stripping metadata removes every AI marker"
It removes the markers stored as metadata. Images made with ChatGPT or Google's models can also carry SynthID in the pixels, which is a separate layer.
"Converting the file to another format removes SynthID"
Converting re-encodes the pixels, but Google describes SynthID as designed to survive common transformations such as compression. A format change is exactly the kind of edit it was built to withstand.
"If the SynthID Detector finds nothing, the image isn't AI-generated"
The detector looks for Google's watermark. Many generators don't use SynthID at all, so a negative result only tells you that this particular watermark wasn't found.
"Removing a watermark is the same as removing metadata"
They are different operations. Removing metadata deletes data stored next to the picture and leaves the picture untouched. Removing an embedded watermark means altering the picture itself.
SynthID beyond images
Google applies SynthID to more than still images: video made with Veo, audio from Lyria and text from Gemini (Google). In video the watermark is carried in the frames, so the same rule applies: removing a video's metadata boxes, as our video metadata remover does, leaves the frames and the watermark exactly as they were.
Check what metadata a file carries with the main tool, or read about C2PA Content Credentials.
Frequently asked questions
Can this website remove SynthID?
No. This tool only removes metadata, and SynthID is embedded in the pixels. We don't offer pixel watermark removal.
Does SynthID show up in a metadata viewer?
No. It is not stored as metadata, so EXIF, XMP or C2PA viewers will not show it. Google's SynthID Detector is the tool built to look for it.
Do ChatGPT images have SynthID?
OpenAI says images made with ChatGPT and its API include both C2PA metadata and SynthID watermarks.
Is SynthID the same as C2PA?
No. C2PA is a signed metadata record attached to the file. SynthID is an invisible watermark inside the content. A file can have one, both or neither.