What Is SynthID — and Can You Remove Invisible AI Watermarks?

Visible logos and metadata are the easy part. The harder case is an invisible watermark — a signal woven directly into the pixels of an image, designed to survive cropping, compression and recolouring. Google’s version is called SynthID. Here’s an honest look at what it is and what removing it actually takes.

What is SynthID?

SynthID is a watermarking technique from Google DeepMind that embeds an imperceptible pattern into AI-generated images (and audio/video). You can’t see it, and it isn’t stored as metadata — it’s part of the pixel data itself, which is what makes it robust. Only Google’s own detector can reliably read it.

Why it’s different from a logo or metadata

• A visible logo sits on top of the image — remove the overlay and it’s gone.
• Metadata (C2PA, EXIF) lives beside the pixels — strip the metadata block and it’s gone, losslessly. See removing C2PA and “Made with AI” metadata.
• An invisible watermark is the pixels. You can’t simply delete it — you have to change the image data itself.

Can you remove an invisible watermark?

It’s possible, but it’s a fundamentally heavier process: the image has to be regenerated so the embedded signal no longer survives, which means the output is a faithful look-alike rather than a byte-for-byte surgical edit. That takes real compute (a GPU), so it’s a premium capability rather than a free, instant one.

On Unmarkr, visible-logo and metadata removal are free and instant today. Invisible-watermark removal is a premium option behind explicit consent — the right tool depends on which kind of mark you’re actually dealing with.