SynthID is Google DeepMind’s watermarking system for identifying content created or edited with Google AI products. For images, Google describes the watermark as invisible to people and embedded in the image. The purpose is to help identify AI-generated content while preserving the image’s visible quality.
That description is useful, but it does not mean every image has SynthID, that every detector can find it, or that a watermarked image can be identified from its metadata alone.
How it differs from metadata
Metadata is information attached to a file in structured fields. It may record a camera, date, location, caption or editing tool. A watermark is information carried in the image content. It is not simply another EXIF or XMP field.
This difference matters when making a clean copy. Removing ordinary metadata may remove details such as a GPS coordinate, but it does not automatically remove a watermark embedded in image pixels. A photo can also have no location metadata and still contain an invisible watermark. The reverse is possible too.
What Google says about robustness
Google DeepMind says SynthID was designed to remain detectable after some common image changes, such as cropping, adding filters, changing colors and lossy compression. That is a description of the system’s design. It is not a promise that every version of every image can always be identified after any edit.
The result of a watermark check should be reported as a check by a particular detector, with its limits. A lack of a detected watermark is not proof that a person made the image. A detected watermark is evidence relevant to how the image was generated or edited; it is not a complete account of every person or tool involved.
Credentials provide a different record
Content Credentials are signed provenance records. The C2PA standard describes both cryptographic bindings to file content and soft bindings that can help associate related renditions. These records have their own verification rules. They are not equivalent to SynthID, and metadata cleanup is not a substitute for validating a credential.
What the deprint preview does
The SynthID step shown on this site is simulated. The preview does not run a SynthID detector and does not remove or weaken a SynthID watermark. It cannot tell you whether an uploaded image was generated or edited with AI. The sample results are visual examples only.
Deprint’s photo workflow focuses on making a separate copy with ordinary metadata removed. Keep the original if its metadata, credentials or history matter to you. Use the copy only for a purpose you have the right to pursue, and do not describe a metadata-cleaned file as proof of human authorship.