What Is SynthID? The Invisible Watermark
Last updated:
SynthID is Google DeepMind’s system for marking AI-generated media so machines can recognize it while people cannot. Here is what it actually does to your files.
A watermark you cannot see
SynthID does not draw anything on the image. Instead it nudges individual pixels by amounts far below the threshold of human perception, distributing a signature across the frequency domain of the picture. Every image from Gemini, Nano Banana, Imagen and Veo carries one, whether you are a free user or a paying subscriber, with no switch to turn it off.
Because the signal lives inside the pixels rather than beside them, it is built to survive cropping, resizing, filters and re-saving. That robustness is deliberate: Google wants AI output to stay identifiable after ordinary editing.
It covers video and audio too
SynthID for video embeds the signal in every frame, which is why per-frame processing is the only honest way to clean a clip. SynthID for Audio does the same in the frequency spectrum of music and speech generated by tools like Lyria.
The detection side stays closed: Google’s official checker is limited-access, so no third-party site can promise certified SynthID detection. Honest tools report C2PA and metadata deterministically and clearly label SynthID findings as inference.
How removal works
Removing SynthID means transforming the pixel data itself — typically a sub-pixel resample plus low-level noise, then re-encoding, which also strips all metadata in the same pass. It cannot be "deleted" the way a tag is; it has to be disrupted.
Use removal on files you generated or own. The point is controlling how your own work is labeled, not deceiving anyone about its nature.
Take the label off your own work
Scan my files free