AI Watermarks · Explainers · August 2026
Gemini Watermark & SynthID: What Google Actually Does
Published: August 17, 2026 · Last updated: August 28, 2026
Google's Gemini watermarking is two different things that often get confused: a visible badge that can now be turned off, and an invisible system called SynthID (plus C2PA metadata) that stays. Here's the difference — and what it means for AI text.
Visible watermark: now optional
Google has made the visible Gemini watermark optional for AI-generated images, video and music. Users can turn the badge off. But that only removes what you can see.
Invisible provenance: SynthID + C2PA
Underneath the visible badge, Google keeps two provenance systems running:
- SynthID — Google's invisible watermarking system, used to identify AI-generated or AI-edited content. Google states that SynthID can survive transformations such as resizing, recoloring, compression and other modifications.
- C2PA / Content Credentials — metadata that records the origin and edit history of an asset. It travels with the file and can be read by tools that support the standard.
That's the key distinction: turning off the visible watermark doesn't remove the invisible provenance. It's why "removing the Gemini watermark" means different things depending on which layer you're talking about.
Can SynthID be removed?
Google designed SynthID to survive common edits — resizing, compression, recoloring. Research into watermark-removal attacks is active, but the honest answer is that invisible, transformation-resistant watermarks are not something a simple tool strips out. Claims of guaranteed SynthID removal should be treated with skepticism.
What about Gemini text?
SynthID is primarily associated with images, audio and video. AI text watermarking is a separate, harder problem — that's where Anthropic's planned Claude watermark (August 2026) sits. Read our Claude watermark explainer →
This is a common point of confusion: people search for a "Gemini text watermark" or a "Gemini watermark remover" expecting the same thing as the image badge. Gemini text output can carry formatting and Unicode artifacts from copy-paste — the same layer ClearPaste cleans for ChatGPT and Claude — but that is not SynthID. Clean Gemini text formatting →
The Gemini watermark layers at a glance
What ClearPaste can clean
For text, the removable layer is the formatting and Unicode one:
- Markdown clutter — headings, bullets, asterisks, code fences
- Invisible Unicode — zero-width spaces, joiners, BOMs, bidi marks (emoji-safe)
- Typography — em/en dashes, smart quotes, ellipses, non-breaking spaces
- Whitespace — trailing spaces, doubled blank lines
Everything runs in your browser. Your text never leaves your device. Open the cleaner →
See what's actually in your AI text
Paste text below and inspect exactly which artifacts are hiding in it.
Open the AI Text CleanerFrequently asked questions
What is SynthID?
SynthID is Google's invisible watermarking system for identifying AI-generated or AI-edited content. Google says it can survive transformations such as resizing, recoloring, compression and other modifications.
Can the Gemini watermark be turned off?
The visible Gemini watermark can be turned off for generated images, video and music. The invisible provenance — SynthID and C2PA metadata — remains.
Can SynthID be removed?
SynthID is designed to survive common edits like resizing and compression. Active research exists on removal attacks, but there's no simple tool that guarantees removal — treat those claims skeptically.
What is C2PA / Content Credentials?
C2PA (Coalition for Content Provenance and Authenticity) is a metadata standard. Content Credentials record an asset's origin and edit history and can be read by supporting tools.
Does ClearPaste remove Gemini watermarks?
ClearPaste does not claim to remove or verify SynthID or C2PA. It cleans formatting and Unicode artifacts in text — the layer that actually causes copy-paste problems.
Does Google watermark Gemini text?
SynthID is primarily associated with images, audio and video. Text watermarking is a separate problem — Anthropic announced a planned invisible text watermark for Claude in August 2026. Read more →