The StripSynthID guide
How SynthID works inside an image.
SynthID embeds an invisible signal in image pixels. File metadata is stored separately, so removing supported metadata does not remove that pixel-level signal.

A watermark is not always a logo.
A visible mark is part of the image people can see. Google describes SynthID as an invisible digital watermark embedded in generated media. Its image watermark is designed to remain detectable through many common transformations, including changes that remove ordinary metadata.
Provenance records are another layer.
A file can also hold camera metadata, editing history and C2PA Content Credentials. These records and the embedded signal serve different roles. Removing a file record is not evidence that a pixel watermark was removed.
Read the detector result precisely.
A positive result supports the presence of a signal recognized by that detector. A negative result means the signal was not detected in that test; it does not establish human authorship. Unsupported formats, unavailable services and failed requests must not be reported as clean results.
Build a repeatable comparison.
Keep the original. Confirm its detector result before processing, then compare the output with the same service and record the date. Inspect image quality separately. A detector result alone cannot tell you whether text, faces or important visual details survived.