Google extends SynthID watermarking to Reimagine edits in Magic Editor

What changed this week in Google Photos

Google Photos is starting to apply SynthID to images edited with Reimagine, a generative feature inside Magic Editor. Until now, the more visible use of SynthID has been on images produced entirely by AI, such as output from Google's text-to-image model Imagen.

The shift here is subtle but concrete: the watermark is moving from wholly synthetic images to partially edited real photographs. That is a different problem. A fully generated image is unambiguously synthetic; a photo where a user regenerated part of a scene sits in a gray zone, and Google is now trying to mark that gray zone.

The detection floor Google is openly admitting

The most important detail in the announcement is a limitation, not a capability. Google states plainly that some Reimagine edits may be too small for SynthID to label and detect.

In some cases, edits made using Reimagine may be too small for SynthID to label and detect — like if you change the color of a small flower in the background of an image.Montana Labs

This means the absence of a SynthID watermark does not prove an image is unedited. There is a floor below which the technique does not register a change. For anyone treating the watermark as a binary trust signal, that caveat matters more than the headline feature.

Why the fallback to 'About this image' is telling

Google points users to its 'About this image' tool, which shows whether a SynthID watermark is present alongside an image's metadata. The pairing is notable: the watermark alone is not being presented as sufficient, so metadata is offered as a companion signal.

That layering acknowledges that a single embedded watermark cannot carry the full transparency burden, especially given the small-edit gap. It pushes verification toward a lookup step rather than something a viewer can read directly from the image.

The implication: a watermark that covers most edits, not all

The practical takeaway is that SynthID on Reimagine is a coverage improvement with an explicit boundary. It captures substantial generative edits but not minor ones, and Google says it will keep gathering feedback and evaluating additional solutions.

For teams building on top of image provenance, the safe reading is that a present watermark is meaningful evidence of an AI edit, but a missing one proves nothing. Detection tooling should treat SynthID as one input among several — metadata, provenance records, context — rather than the final verdict on whether an image was touched by generative AI.