Google's SynthID watermark proves resilient in extensive image‑edit tests

Google’s SynthID technology embeds an invisible watermark directly into the pixels of AI‑generated images and videos, allowing the content to be identified even after typical online transformations. The company announced that the tool has already been adopted by partners such as OpenAI, Runway and Nvidia, and claims the watermark survives compression, scaling, cropping and other edits.

An independent practical test conducted in London showed that SynthID detection remained stable after hundreds of compression and save cycles. The watermark began to fail only when image edges were heavily cropped or when severe data loss occurred. The study also contrasted SynthID’s hidden watermark with the cryptographic metadata standard C2PA, noting that while C2PA can be stripped more easily, SynthID offers a more covert but robust labeling method. The findings highlight both the promise and the limits of watermarking as a tool against AI‑generated misinformation.