Anyone can now check your clip for Google's AI watermark. What SynthID Detector means for clippers.

On 6 October 2026 Google opened SynthID Detector to anyone with an account.

Upload an image, a video or an audio file, and it tells you whether the file carries SynthID, the invisible watermark Google puts in its AI output. The portal's front page lists four companies whose generated media it can detect: Google, Nvidia, OpenAI and Kakao.

On 8 October, Google's autocomplete finished "synthid" with "synthid detector". It also offered "synthid remover".

That second search is the one that worries me.

Here is the complication. A clipper who drops a Veo shot or a generated background into a campaign clip used to assume nobody could tell. Now the brand, the campaign reviewer and the viewer in the comments can all upload the same file and check.

Check 1: What the detector reads

The portal's FAQ says SynthID is "woven into the content itself: into the pixels for images and video and the frequency components of audio."

That matters because it is not metadata. Re-saving, screenshotting or re-exporting does not strip it the way it strips a file tag.

The FAQ also says the watermark is built to survive filters, colour changes and lossy compression.

For video and audio, the portal flags the segments that carry the mark. So a two second generated insert inside a forty second clip can be located, rather than suspected.

Check 2: What a clean result does not prove

The portal is blunt: "This is not a general AI detector."

It only knows about media from the companies that adopted SynthID. A file from any other model comes back clean.

The FAQ lists the reasons a generated file can return "Watermark not detected": it came from a model without SynthID, or heavy modification degraded the signal. It also says that in very rare cases the system can falsely trigger on content that was never watermarked.

So two rules.

A hit is strong evidence that a watermarked model made or edited the file. A miss is no evidence that a human did.

If a brand or a campaign manager tells you your clip "passed" or "failed" an AI check, ask which tool and which result. A clean SynthID scan does not clear anyone.

Check 3: Why a SynthID remover is a bad trade

The contrarian line here is short. Do not strip it.

Three reasons.

First, the portal's own terms, as PPC Land reported them on 7 October 2026, bar using the service to optimise edits designed to break or bypass detection. Testing a file over and over until it comes back clean is the exact thing those terms describe.

Second, the FAQ admits that someone who tries many transformations may find one that defeats the mark. That is a reason the remover tools exist. It is also a reason a clean scan carries no weight, so stripping buys you nothing a reviewer will trust.

Third, YouTube does not rely on SynthID alone. Its help page on disclosing generative AI content says it may label content made with its own AI tools, content carrying C2PA metadata, and content its systems detect as AI generated or altered. Labels from C2PA metadata or manual review "can not be adjusted" by the creator.

And creators who consistently choose not to disclose may get a label applied manually, or penalties up to removal or suspension from the Partner Program, in YouTube's words.

A clipper's account is the asset. Gambling it to hide a three second b-roll shot is a bad trade.

Check 4: What needs disclosing

Most clipping work never touches this.

YouTube's disclosure page lists what does not need a label: colour and lighting filters, captions, upscaling and repair, background blur, using AI for a script, title or thumbnail, and cloning your own voice for dubs.

What does need disclosure is realistic content that could mislead: making a real person appear to say or do something they did not, altering footage of a real event or place, or generating a realistic scene that did not happen.

For campaign clips that usually means three cases.

1. Generated footage of a real place used as if it were filmed there.
2. A generated shot of the speaker, or anyone real, doing something they did not do.
3. AI music that carries the clip. YouTube's page lists AI-generated music that is the main focus of the video as something to disclose.

A stylised generated background behind a podcast host is a grey area. Disclose it anyway if it looks real.

Check 5: What to tell a campaign before you post

Whether or not the brief mentions AI, put it in your submission note, every time a clip uses generated footage, voice or music:

• which seconds are generated
• which tool made them
• whether you ticked the AI use box on upload

That note costs ten seconds. It turns "your clip failed an AI check" into "we already knew, it was in the note."

VALORAE Cast routes members toward paid campaign work, starting from $0.80 per 1,000 views, and every rate depends on a clip the brand is willing to keep paying for. A clip that gets pulled for undisclosed AI pays nothing, however many views it had.

Check 6: Use the detector on your own sources

The portal is also useful the other way round.

If you pull a clip of a public moment from a repost account, run it before you cut. A hit on footage presented as real is a reason to find the original or skip it. Clipping a generated "real event" puts your account on the wrong side of the same disclosure rules.

The portal's terms bar bulk automated checking, per PPC Land, so this is a spot check on footage you doubt, not a pipeline.

And keep the original file you checked. If a brand later questions a clip, the untouched source and your note on when you checked it are the evidence you will want.

VALORAE Media went through the monetisation side of this for business channels in can AI video get your channel demonetised. The rules a business channel reads are the same ones a clipper's account lives under.

The short version

• SynthID Detector at synthid.com is open to anyone signed in since 6 October 2026.
• It detects watermarks from Google, Nvidia, OpenAI and Kakao only.
• A hit is strong evidence. A clean result proves nothing.
• Do not use a SynthID remover. Disclose instead.
• Tick YouTube's AI use box for realistic generated footage, people, places or music.
• Note generated seconds and tools in every campaign submission.
• Spot check doubtful source footage before you cut it.

VALORAE Cast is the Discord community for editors and clippers at discord.gg/valoraecast, where questions like this get asked before they cost anyone a payout.