SynthID Detector: how to check a video for a watermark today
Google DeepMind's SynthID page says a detector portal exists with an early-tester waitlist, and Gemini can check uploads. What that means for a video team.
You can check content for a SynthID mark in two ways named on the Google DeepMind SynthID page: the SynthID Detector portal, which has an early-tester waitlist, and uploading the content in Gemini and asking. The page says Google is working with journalists and media professionals to test the portal.
What SynthID covers per the page
The page lists images and video, audio such as Lyria and NotebookLM podcasts, and recently Gemini text. It describes a watermark that is tested with a detector, not a label a viewer sees.
| Item | Page says |
|---|---|
| Media | Images, video, audio, and recently Gemini text |
| Portal | SynthID Detector, early-tester waitlist |
| Quick check | Upload in Gemini and ask |
| Testers | Journalists and media professionals |
What it cannot tell you
A detector answers for its own watermark. A clip from another vendor has no reason to carry a SynthID mark, so a negative result is not proof a clip is real. Likewise the Sume docs do not say Sume output is watermarked, so do not expect a hit on a Sume clip.
Practical use
If you publish AI video, a detector check is one input to an evidence file. Pair it with your own job record. See the evidence log pattern and what Sume does and does not add to video.
Sources
Related posts
More in Use cases
- A 10-minute avatar module on Sume: ten jobs, priced by tier
Sume avatar videos run 4 to 60 seconds, so a 10-minute module is ten jobs and a join. Standard comes to $110.40 and plus to $147.00, by arithmetic from list.
- Tennessee ELVIS Act and AI voice: a checklist for TTS projects
Tennessee's ELVIS Act, signed 21 March 2024, extends protection to voice. A consent checklist for AI voiceover, using the governor's page and Sume's TTS docs.
- Text edited in an ad image: a dimension check before you ship
After an AI text replacement, compare output and source dimensions in Pillow. Stretched or recropped edits are caught in one assert, before the ad goes live.
- Text to speech from a PDF with an API: extract, chunk, narrate
Sume's TTS takes text, not PDF files. Extract the text, split it under 20,000 characters, narrate each chunk, then join the audio. Costs and limits included.
Written by Sume