SynthID Detector Is Free: What It Catches & Misses
On October 7, 2026, Google made its SynthID Detector free for everyone at synthid.com. Until then it had been limited to researchers and journalists after its Google I/O preview. Searches for "SynthID detector" and "is it real or AI" have jumped since the launch, so here is a plain-English guide to what it does and where it falls short.
What SynthID Detector checks
SynthID is an invisible watermark woven into the pixels, frames or audio that Google's models produce. The public detector accepts images (JPG, PNG, WebP), video (MP4, MOV, WebM) and audio (WAV, MP3, AAC) and reports whether that watermark is present. For video and audio it highlights which segments carry it.
Which AI tools it can see
- Google's own generators: Nano Banana, Veo, Lyria, the Gemini app, Flow and Google Vids
- Partners that adopted SynthID, including OpenAI image and audio output, NVIDIA and Kakao
- Apple has said support is coming
What it misses
Google itself states that it is not a general AI detector. It returns "not detected" for anything without its watermark, including:
- Midjourney images
- FLUX and other open models run locally, such as Stable Diffusion
- Runway, Kling and other video tools that use their own marking
- Text of any kind
- Media that has been heavily edited, which can weaken the watermark
How to fill the gaps
Many of the tools SynthID can't see still leave traces in the file's metadata: IPTC AI tags (used by Midjourney), Stable Diffusion and ComfyUI generation parameters, and C2PA Content Credentials (used by Adobe, Microsoft and OpenAI). Our free AI image detector reads all of these in your browser.
The best workflow in 2026
- Get the original file, not a screenshot.
- Run it through SynthID for Google and partner watermarks.
- Run it through AIScan Lens for metadata and Content Credentials.
- If both find nothing, check the source and do a reverse image search. "Not detected" is never proof that something is real.