Skip to content
Unmarker
Explainer and tool

SynthID-Text watermark: detector, remover and how it works

SynthID-Text is Google DeepMind's statistical watermark for text, published in Nature in 2024 and active in Gemini. Anthropic also marks text from its newer Claude models, without having published its method. Here is how the mark gets into the words, how a detector reads it, and what is known about weakening it.

Test AI watermark removal

5 € pack: 20,000 words, no subscription

Clear rewrites in your own voice.Updated
Key facts
  • Whole-sentence rewriting that fits the context
  • Changed words visible at a glance
  • Exact numbers, URLs and protected phrases preserved
  • English, French, Spanish and German, 10 to 200 words per request
  • Free: 300 words per day, maximum 3 attempts
  • 5 EUR pack for 20,000 words, one-time payment, no subscription
  • No statistical watermark detector or score, and watermark removal is not verified

1. Paste your text

Enter 10 to 200 words in English, French, Spanish or German. Exact numbers, URLs and protected phrases stay untouched.

2. Generate variants

Unmarker Studio rewrites your text in context, in the same language, at the intensity you choose. Changed words are highlighted.

3. Check and review

Variants are prepared using AI, then the final wording is assembled automatically. You review the result and keep control of the final text.

SynthID-Text watermark: how it works and how it is detected

What SynthID-Text is

SynthID is Google DeepMind's family of watermarks for AI-generated images, audio, video and text. The text version, SynthID-Text, was described in a paper published in Nature in October 2024 and released as open source at the same time, so that other developers could apply it to their own models. Google says it has been active in Gemini products since 2024, and the paper reports a live test on Gemini traffic. Anthropic has also said it marks text from its newer Claude models on supported surfaces, but it has not published its marking algorithm, so this page does not treat that mark as SynthID. If you are looking for the Claude specifics, see the Claude watermark remover; this page covers the technique itself: how the mark is added, how it is detected, and what is known about weakening it.

Is Gemini text watermarked? (gemini watermark text)

Google says text generated by Gemini models carries SynthID-Text, Google DeepMind's statistical watermark, deployed since 2024 and described in Nature in October 2024. It is not a hidden character and not metadata: the mark is in the choice of words, and only a detector holding Google's key can read it with certainty. Google's SynthID Detector portal exists for content made with Google tools. Copy-paste generally keeps the mark; light edits can keep much of it; deep paraphrase and translation tend to weaken it. Short passages give little signal either way. For ChatGPT text, OpenAI's public provenance documentation does not establish a statistical text mark, so a "Gemini watermark" question and a "ChatGPT watermark" question have different answers.

How the mark gets into the words

A language model produces text one token at a time by sampling from a probability distribution. SynthID-Text does not add anything to the output. Instead, at each step it uses a secret key and the previous few tokens to compute a pseudo-random score for every candidate token, then nudges the sampling toward tokens with a higher score. Google's implementation uses a procedure called tournament sampling: candidates compete in rounds, and the winner is chosen. Over a whole paragraph, the text ends up slightly enriched in "favoured" tokens. A detector holding the same key recomputes the scores and checks whether the text is enriched more than chance would allow. That is why the mark survives copy-paste and light edits: most of the favoured choices are still there.

Two consequences matter for anyone trying to remove it. First, low-entropy text (a list of dates, a piece of code, a very factual sentence where only one word fits) barely carries the mark, because there was little choice to bias. Second, the mark is only as strong as the number of choices it could bias: short texts give weak, uncertain scores. Google's own material says the watermark is robust to cropping and light modification, and weaker under heavy paraphrase and translation.

How SynthID-Text is detected

Detection is the mirror of marking. A detector that holds the key recomputes, at each position, the pseudo-random scores of the candidate tokens and checks whether the tokens actually present are enriched in high scorers. The Nature paper reports the method's operating points: on long, high-entropy text the mark is found reliably at very low false-positive rates; on short or low-entropy text the power drops, which is why short texts give uncertain results. Google holds the key for Gemini and offers verification through its own products; in May 2025 it announced a SynthID Detector portal for content made with Google's tools. The open-source release lets a developer who applied SynthID-Text to their own model run the matching detector, with their own key. Nobody without the key can be certain. Anthropic says its detection API is in private preview, and its public file checker does not check text.

What can weaken SynthID-Text

Unmarker is a contextual rewriting tool for English, French, Spanish and German. An AI model prepares alternative phrasings of the text in its source language, at a light, balanced or strong intensity, and the final wording is then assembled automatically, so the wording changes while the meaning is meant to stay. Reread the result. Exact numbers, URLs, quoted passages and phrases you protect are kept as they are.

Unmarker has no detector and gives no watermark score. Whether this kind of rewriting removes a statistical watermark has not been verified, and we claim no removal rate and no human or undetectable verdict. Heavy rewriting is expected to weaken the signal, but that is not a guarantee for any given text. The rewriting itself is done by an AI model whose output may carry its own mark; that is an inference from Anthropic's published marking coverage, not a measured result.

Which texts may carry a statistical mark

  • Text produced by Gemini models and Google products that use SynthID-Text.
  • Text produced by new Claude models launched in the EU since August 2, 2026, on the surfaces Anthropic says are supported. Not every historic Claude model or every text is marked.
  • Text produced by any open model to which a developer has applied the open-source SynthID-Text implementation.
  • Human text that was proofread, translated or reformatted by one of those models. Anthropic notes that this can be enough to trigger detection; the same logic applies to any SynthID-style scheme.

Not covered here: images, audio and video. SynthID marks those in the pixel or waveform domain and this page, like Unmarker, is about text. If you searched for "remove SynthID" with a picture in mind, we cannot help, and we would rather say so than sell you the wrong thing. Not covered either: ChatGPT text, for which OpenAI's public documentation does not establish a statistical text mark, and which is handled by the ChatGPT watermark remover (invisible characters and habits). For the status of every vendor with sources, see AI text watermarks compared.

Why manual editing, synonym tools and translation are unreliable

Editing a few words leaves most of the enriched choices in place. Synonym-swap paraphrasers change words one at a time without regenerating the sentence, which lowers the score only partially and often mangles meaning. A translation round-trip (to another language and back) does re-sample the wording, but it also re-samples your meaning, and if you use Gemini or Claude for the round-trip you may simply add a fresh mark. Unmarker works in the same language (English, French, Spanish or German) and rewrites the text in context, with variants prepared using AI and the final wording assembled automatically. The guide on removing the Claude watermark compares these options in a table.

What we will not claim

No tool can promise to remove a statistical watermark. Detection is statistical, keys can change, and short or very constrained texts cannot be rewritten without changing them. We also do not market the tool for cheating: if you owe someone a disclosure, make it. Our note on whether you should explains how we see it.

Sources

Google DeepMind, SynthID (technology page, text section). Dathathri et al., Scalable watermarking for identifying large language model outputs, Nature, October 2024. Anthropic's August 2026 announcement and press coverage are quoted in our Claude watermark guide.

Pricing

Free: 300 words per day, with a maximum of 3 free attempts. Each request takes 10 to 200 words (12,000 characters maximum). Paid: 5 EUR in total, taxes included, for 20,000 words, as a one-time purchase with no subscription. Credits are kept behind a secret recovery key stored in your browser; download the key to recover them on another device.

Questions people ask

Is Claude's watermark the same as Google's SynthID-Text?

This is not confirmed. Anthropic has not published its marking algorithm, so we cannot say it is SynthID-Text. Unmarker does not treat them differently: it rewrites wording in English, French, Spanish or German, and whether that weakens a statistical mark has not been verified.

Can you detect SynthID without Google's key?

Not with certainty. Unmarker does not include a detector. Anthropic says its detection API is in private preview, and its public file checker does not check text.

Does the tool remove SynthID from images?

No. This is a text tool. SynthID for images, audio and video is a different mark and we do not address it.

Will translation remove SynthID-Text?

Partly, and at the cost of meaning. If the translator is itself a watermarked model, you may add a new mark. Rewriting in the same language avoids the translation step and is what the tool does, though the rewriting model may itself add a mark and the effect on an existing watermark has not been verified.

Try it free

Try it free

Clear rewrites in your own voice.

Test AI watermark removal

5 € pack: 20,000 words, no subscription