Google opens public SynthID Detector for checking AI-generated images, video and audio

SynthID Detector is now open to the public. Users can upload images, video or audio to synthid.com and check for invisible SynthID watermarks. The notable change is cross-vendor coverage: besides Google's own models, the detector recognizes SynthID marks embedded by OpenAI, NVIDIA and Kakao, which have adopted the watermarking scheme.

SynthID embeds imperceptible signals in generated pixels, audio samples or tokens that survive common edits like cropping and compression. A detector only helps if many generators watermark, so multi-vendor adoption is what turns this from a Google feature into something closer to an industry standard. It complements C2PA content credentials, which rely on metadata that is easily stripped.

The launch arrives amid rising pressure on provenance from elections, scams and deepfake fraud, and as regulations like the EU AI Act require labeling of synthetic media. Competitors such as Meta and Adobe push their own approaches, and fragmentation remains the main obstacle.

Limits: the detector only finds content from participating generators. Open-weight image models and non-adopting vendors produce nothing to detect, so a negative result does not mean 'human-made'. Watch for adoption by Anthropic, Meta and Midjourney, and for adversarial watermark-removal research.