Research note
Research note: Google opens its SynthID Detector to everyone to check if images, video or audio were made with AI
How we checked the SynthID Detector launch, whose watermarks it reads, and what a “not found” result does and does not mean.
What we set out to check
When the detector opened, to whom, whose watermarks it can read, which files it accepts, and its limits.
What we found
- Google’s post confirms global availability in English from 7 October and names OpenAI, Nvidia and Kakao as partners, with Apple coming soon.
- TechCrunch lists the supported file formats.
Numbers that differ
None. Both sources give the same figure of over 1 million daily checks.
Labels we chose and why
- Confirmed for availability and partners.
- Company claim for scale figures such as 180 billion watermarked items.
- Reported for the limit that it reads only SynthID marks, which follows from both sources.
What we could not verify
- File size or length limits, which neither source gives.
- How often the detector misses a mark after a file is edited.
What we searched
- blog.google SynthID verification website everyone
- SynthID detector OpenAI Nvidia Kakao
- synthid.com supported formats
Every source we read
| Source | Used? | Why |
|---|---|---|
| Google blog | USED | Official post by Google DeepMind. Used for availability, partners and scale figures. |
| TechCrunch | USED | Independent report. Used for file formats, Google models that add SynthID, and limits. |
| FourWeekMBA | NOT USED | Repeats the announcement without new facts. |