Skip to content
Hidden Character Inspector

Private browser-local Unicode inspection

A different mechanism

SynthID Text is not hidden Unicode

SynthID Text is described by Google DeepMind as a watermark introduced through token selection during generation and detected with a scheme-specific statistical process. Deleting zero-width characters is not the inverse of that process, so this site neither detects nor removes SynthID.

Token selection versus literal characters

A language model produces text by selecting tokens from a probability distribution. A statistical watermark can influence that selection pattern without adding a special visible or invisible character to the final string. Detection then evaluates a statistical signal using knowledge of the scheme.

A Unicode inspector answers a simpler factual question: which scalar values are present? It can locate U+200B because U+200B is literally in the input. It cannot infer or reverse a generation-time token-selection pattern from that character inventory.

Why this tool does not offer a workaround

Attempting to weaken a statistical signal would require rewriting content and making claims about a detector this property does not operate. That would be a different product, evaluation corpus, privacy boundary and purpose. It could also change meaning and still provide no reliable guarantee about another system.

The workbench therefore performs no paraphrasing, synonym replacement or model call. It does not label text as watermarked, unwatermarked, AI-written or human-written. Its completed result is limited to the hidden-character policy it actually ran.

How to read a clean result

“No removable hidden-character issues found” means exactly that: under the declared Unicode policy, the inspector found no occurrence it can offer for cleanup. Statistical watermarks and AI authorship were not inspected. For imported files, unsupported container metadata was not inspected either.

That qualification remains visible because a result should not expand into a stronger claim merely because its narrow check passed. If your question is specifically about SynthID, consult the scheme owner’s documentation rather than treating an invisible-character scan as a substitute.

Primary references

These sources describe the external standards or mechanisms discussed above. The property’s cleanup decisions remain narrower than those specifications.

Scope reminder: This workbench inspects hidden Unicode and copy/paste artifacts. It does not determine AI authorship, remove SynthID or strip Content Credentials.