Google's SynthID Detector: What Creators Need to Know

What just launched

On October 7, 2026, Google opened its SynthID Detector to the general public. The tool had existed in a limited form since last year, available only to journalists, media professionals, and researchers through a waitlist. Now anyone can go to synthid.com and upload a file to check for an invisible AI watermark.

The detector checks images (JPG, JPEG, PNG, BMP, WEBP, AVIF, HEIC, HEIF, TIFF, TIF, GIF), video (MP4, MOV, WEBM), and audio (WAV, MP3, OGG, FLAC, AAC, M4A) for SynthID watermarks embedded by Google's own AI models as well as partner models from OpenAI, Nvidia, and Kakao, with Apple joining soon. Google's Pushmeet Kohli says the company has watermarked more than 180 billion images and videos, plus 240,000 years of audio, since SynthID launched in 2023.

This isn't a brand-new technology so much as a new level of access. The underlying watermarking has been running quietly for years inside Google Search, the Gemini app, and Chrome, which together already handle more than one million verification requests daily, according to Google. What changed on October 7 is that the checking tool is now in everyone's hands, not just newsroom fact-checkers.

Why this matters for short-form creators

Up to now, whether a clip was AI-generated or AI-edited was mostly something a creator decided to share or not. That's shifting. Viewers, brands doing sponsorship vetting, and platforms can now run a one-click check on any video, image, or audio clip a creator posts, without needing special access or technical skill.

Google also says SynthID watermarks survive cropping, filters, compression, and speed changes. That matters for short-form specifically, because this is exactly the kind of editing nearly every clip goes through before it's published: trimmed, re-cropped for vertical formats, run through filters, sped up or slowed down for pacing. A watermark that only worked on untouched files would be close to useless for how creators actually edit. One that survives typical editing means AI involvement can still be traced well after a clip has been reshaped into a final post.

The practical effect: creators who use AI tools and say nothing are increasingly likely to get checked, not because anyone is hunting for them specifically, but because checking is now trivially easy for anyone curious enough to do it.

What the detector can't do, and why that matters too

It's worth being precise about the tool's limits, because they cut both ways for creators. The detector does not prove a file is human-made or authentic. It only tells you whether it found a SynthID watermark from a participating partner. If a file returns no match, that could mean the content is genuinely untouched by AI, or it could mean the watermark simply wasn't detected, or that it was made with a tool outside the partnership.

That partnership list is the key limitation. The detector catches watermarks from Google, OpenAI, Nvidia, and Kakao, with Apple coming soon. It does not detect Meta's Content Seal, and it isn't a general-purpose AI classifier that works across every model worldwide. Content from many Chinese models and various open-source image generators can pass through undetected simply because those tools aren't part of the SynthID partnership.

This means a 'clean' result from the detector isn't proof of anything definitive, and creators using tools outside the Google/OpenAI/Nvidia/Kakao ecosystem shouldn't read a no-match result as a guarantee of invisibility. The tool's coverage will likely expand over time, and relying on gaps in today's coverage is a short-term bet at best.

The smart move: disclose before you're checked

Given that verification is moving from a niche capability to a one-click public habit, the creators who come out ahead are the ones who disclose AI involvement proactively rather than waiting to be caught by a watermark scan. Getting flagged by a viewer or brand after the fact reads very differently than a creator who labeled their process from the start.

This is also increasingly a legal expectation, not just a reputational one. Since August 2, 2026, Article 50 of the EU's AI Act requires providers of generative AI systems to mark AI-generated images, video, audio, and text in machine-readable form, and requires anyone publishing deepfakes to disclose them. Disclosure habits that creators build now will likely be required practice later, in more places.

• Name the specific tools used (for example, a particular AI video or voice generator) rather than a vague 'AI was involved' note
• Flag which shots, voiceovers, or visuals are AI-generated versus filmed or recorded normally
• Use consistent on-screen labeling or caption language across every video, not just the ones you think might get scrutinized
• Put the same disclosure in the video description for anyone who checks after the fact

Building disclosure into your workflow

The easiest way to make this sustainable is to treat disclosure as a standard production step, not an afterthought you add when a video feels risky. Decide on a consistent caption format or label style now, before audience-side verification tools are the default expectation rather than a novelty.

For creators managing scripts, captions, and video descriptions across multiple platforms, this is a good moment to formalize that step in whatever production workflow or tool you use, Episodo included, so disclosure language goes out consistently with every upload instead of being something you remember occasionally.

None of this requires overhauling how you use AI tools. It just means being upfront about it in the places viewers and brands already look: captions, descriptions, and on-screen text. That's a far better position to be in than explaining yourself after someone runs your clip through a detector.