What is SynthID? An honest explanation of Google's invisible AI watermark

2026-10-16 · 4 min read

People often mix up three different things on an AI image: the visible sparkle, the hidden file data and an invisible watermark. They are not the same, and tools treat them differently. This page explains SynthID in plain words.

Three different things

What	Where it is	Can you see it?
Visible sparkle (logo)	Drawn over the corner of some Gemini images	Yes
Metadata (EXIF, XMP, content credentials)	Stored in the file next to the picture	Only with a viewer
SynthID	Woven into the pixels of the picture itself	No

What SynthID is

SynthID is a technology from Google DeepMind. It adds a pattern to the pixels of images made with Google’s AI tools. The pattern is not visible to people. A detector can look for it and say that the image was probably made with Google AI. Google says it is designed to stay detectable after common changes such as resizing, compression or small edits.

Why it exists

As AI pictures become more realistic, it is useful for platforms, companies and newsrooms to be able to tell whether an image was machine-made. A watermark that lives in the pixels is harder to lose by accident than a note in the file.

What Image Cleaner does and does not do

It removes the visible sparkle and the metadata.
It does not touch SynthID. We do not remove it and we will not add a feature to do so. The tool is made for people who are open with clients that an image is AI-generated.

What this means for you

A picture can look clean and still be recognised as AI by a detector. That is expected.
If someone you work with uses an automated AI check, tell them the images are AI-generated and ask them to approve or exempt them. See the disclosure checklist.
If you need images that do not carry this kind of watermark, choose a tool that does not add one, and say which tool you used.