Skip to content

SynthID Text · documented

Gemini Watermark Remover for Text

Google documents SynthID Text as its watermark for generated text. It shapes which words Gemini picks and adds no characters, so there is nothing to find and delete. Simple Unmark rewrites the passage so those word choices change, and removes any hidden characters the text picked up while being copied.

Privacy mode

Google Gemini: current status

· Documented by the provider

Google documents SynthID Text as its method for watermarking AI-generated text.

Status
In production
Mechanism
A logits processor applied during generation. Google describes it as augmenting the model's logits with a pseudorandom g-function, so the watermark lives in which tokens were chosen rather than in any added symbol.
Hidden characters involved?
No. SynthID Text does not insert hidden characters; nothing is appended to the visible text.
Public detector
Google publishes the watermarking and detection approach and has open-sourced a reference implementation, but a public detector for arbitrary Gemini text is not offered as a general consumer service.

What the provider says weakens the signal

  • Google states detector confidence "can be greatly reduced when an AI-generated text is thoroughly rewritten"
  • Translation into another language
  • Factual responses, where there is less opportunity to augment generation
  • Short excerpts, though the method is robust to cropping and mild paraphrasing

How this works

Deterministic marks can be removed exactly. Probabilistic patterns can only be reduced. Simple Unmark keeps the two separate and does not promise detector outcomes.

  • SynthID Text runs during generation as a logits processor. Google describes it as augmenting the model's logits with a pseudorandom g-function, so the watermark is a property of which words were chosen.
  • That is why searching Gemini output for a hidden symbol finds nothing, and why removing invisible characters does nothing to the SynthID signal. Characters you do find usually came from the page, editor or app you copied through.
  • Google says detector confidence can be greatly reduced when text is thoroughly rewritten or translated, and that watermarking is less effective on factual responses. It also describes SynthID as robust to light changes such as modifying a few words or mild paraphrasing.

Three steps

  1. 1

    Paste the Gemini text. The browser scan shows whether the copy also picked up invisible characters from the page or editor.

  2. 2

    Run the clean. Hidden characters are removed, then the whole passage is rewritten so the original word choices change.

  3. 3

    Read the rewrite against the original, check facts and figures, then copy it.

Worked example

Before

Remote work has changed how companies hire. Firms can now recruit from a much wider talent pool, and many report lower office costs. However, managers say onboarding new staff remotely remains a challenge.

After

Hiring looks different once a team goes remote. A company is no longer limited to people who live near its office, and plenty of firms now spend less on rent. The hard part, according to managers, is getting new hires up to speed without sharing a room.

Illustrative editing example written by us, not saved tool output. It shows the scale of change a thorough rewrite makes: new sentence order, new structure, nearly every word replaced. It was not checked with a SynthID detector, and real rewrites vary per request.

Scope

What this page can and cannot do

Can

  • Rewrite the whole passage so the word choices SynthID shaped are replaced, which is the change Google says reduces detector confidence
  • Remove invisible characters the text picked up from the page, editor or CMS it passed through
  • Show what Google documents and when we last checked it, in the status block above

Cannot

  • Delete SynthID Text like metadata, because there is no character or field to strip
  • Report a SynthID score, because Google does not offer a public detector for arbitrary text
  • Guarantee a result, since Google itself describes reduced confidence rather than removal
  • Touch SynthID in images, audio or video. Those are separate systems, and Simple Unmark handles text only

We don’t store submitted content or cleaned results in our application database, and we don’t send content to analytics. For rewrites, Private mode routes plaintext through the app, its load balancer and the processors. Confidential mode encrypts text from the browser to the workload, so the web backend only receives metadata. DeepInfra still rewrites the text. The deterministic scan in Invisible character remover runs entirely in your browser. See Privacy Policy and Methodology.

Questions

About Gemini Watermark Remover for Text

Sources

Does Gemini watermark text?
Google documents SynthID Text as its approach to watermarking AI-generated text and has described deployment at scale in Gemini, along with an open-source reference implementation. See the dated status block on this page for what is first-party documented versus reported.
Does Gemini add hidden characters to text?
Not as a watermark. SynthID Text adds no characters. If a scan finds zero-width spaces or no-break spaces in Gemini text, they most likely came from the chat interface, a web page or an editor along the copy path. The free invisible character remover lists each one by code point.
Can a SynthID text watermark be deleted like metadata?
No. The pattern is part of the wording. There is no field or hidden character to strip, so weakening it takes genuinely new phrasing.
Can I check whether the watermark is gone?
Not with this tool, and not with any public tool we know of for arbitrary Gemini text. Google publishes the detection approach but not a general consumer detector, so be sceptical of anyone who offers to verify SynthID removal for you.
Does this work on Gemini-generated images?
No. Simple Unmark handles text only. SynthID in Gemini images, audio and video is embedded in the media data itself, and this tool does not touch it.