On October 7, 2026, Google made synthid.com available to the public. You can now upload an image, video, or audio file to check for an invisible SynthID watermark, which Google and a few other companies embed into their AI-generated content. Previously, access to this detector was granted to journalists and researchers by application only.
For short-form video creators, this is far from abstract news. TikTok, YouTube, and Instagram require creators to label realistic AI content and also scan for it using technical analysis. If a video is AI-generated but unlabeled, the platform might forcibly add a label, limit its reach, delete the video, or issue a warning. Repeat offenses can lead to demonetization. The SynthID detector reveals what these systems see in your files. Let's break down what it finds, what it misses, and why 'not detected' doesn't mean 'human-made.'
What Is SynthID and What Launched on October 7?
SynthID is an invisible watermark. When a Google AI model generates an image, video, or voice, it simultaneously alters the file in a way that's imperceptible to humans but readable by a program. The technology has been in use since 2023. According to Google, over 180 billion images and videos and 240,000 years of audio have been watermarked.
Google has already integrated this watermark check into Search, the Gemini app, and Chrome, which together handle over a million queries a day, according to the company. On October 7, a standalone site was added to the mix: synthid.com, available globally and currently in English only.
What the Detector Checks For
Crucially, the detector doesn't look for 'traces of AI in general' but for the specific SynthID watermark. That's why the list of supported sources is more important than the list of file formats.
| Whose Content | Status |
|---|---|
| Google AI (Gemini, image, video, and audio generation) | Detected |
| OpenAI | Detected |
| NVIDIA | Detected |
| Kakao | Detected |
| Apple Intelligence | Coming soon |
| All other generators | Not detected |
Formats accepted by the site (according to Unite.AI):
| Type | Formats |
|---|---|
| Images | JPG, PNG, WEBP, HEIC, and others |
| Video | MP4, MOV, WEBM |
| Audio | WAV, MP3, OGG, FLAC, AAC, M4A |
The result is simple: either the watermark is detected or it's not. According to FourWeekMBA, the public version of the detector doesn't show where exactly in the file the watermark is located—that detailed map was part of the internal version for researchers. The publication also notes that you need a Google, OpenAI, or Apple account to log in, and the free tier allows for about ten checks per person per day.
Why the Watermark Survives Editing
This is the part creators usually underestimate. SynthID isn't stored in the file's metadata, which gets wiped with the first re-upload. The watermark is embedded into the image or audio itself, so it persists after:
- cropping the frame or resizing for vertical video;
- applying filters and color correction;
- changing the frame rate;
- lossy compression—including the kind platforms apply during upload;
- for audio—adding noise, recompressing to MP3, and changing the speed.
Simply put: if you take a Gemini-generated background, add your own voiceover, edit it, and post it to Shorts—the watermark in that background will most likely remain detectable.
What the Detector Can't Do
Google is clear that this is not a universal AI detector. Here are its limitations:
- The absence of a watermark proves nothing. If a generator isn't part of the SynthID initiative (and most aren't), its files will pass the check as 'clean.' A negative result does not mean the content was shot on camera.
- It can't distinguish between 'fully generated' and 'lightly edited.' The watermark is either in the file or it isn't. The detector doesn't care if it was a fully AI-generated scene or your real video where AI just touched up the sky.
- It doesn't identify the specific model used. The result is about the presence of a watermark, not its origin.
- Screenshots and re-photographing can disrupt the watermark. The mark is resilient to file processing but isn't legally binding proof.
Alongside SynthID, there's the C2PA (Content Credentials) industry standard—this is an open, visible record in the file about how it was made. The invisible watermark and this record complement each other: the record is easy to remove but contains details, while the watermark lacks details but survives editing.
What This Means for Short-Form Video Creators
Platforms don't label AI content out of the goodness of their hearts; they do it to enforce their rules. TikTok, YouTube, and Meta require creators to label realistic generated video themselves, and their systems also scan files for technical signs of generation to apply labels automatically. An AI label itself won't harm your video. What's harmful is the lack of one: if a creator fails to add a label and the platform detects AI, the video's reach can be limited, it might be deleted, the account gets a warning, and repeat violations can lead to demonetization.
Here's what that means in practice:
1. Know Your Generators
Divide your tools into two groups: those that apply SynthID (everything from Google, OpenAI, NVIDIA, and Kakao) and all the others. This isn't a reason to stop using the first group—it's a reason not to pretend your video was shot on a camera.
2. Label Your Content Yourself
TikTok, YouTube, and Instagram have an 'Made with AI' toggle. Using it yourself costs nothing, whereas having a platform apply a label for you retroactively looks like an attempt to deceive—and on YouTube, that label stays on the video's info card forever.
3. Check Content You Get from Others
This is the detector's most practical everyday use. Found a 'real' clip of an event for your compilation, downloaded someone else's stock footage, or received source files from a client? Run them through synthid.com. If a watermark is found, you'll know what you're working with before you publish.
4. For Brands: Vet Your Deliverables
If you're buying UGC videos, the detector is a cheap filter at the gate: the creator promised a live-action shoot, but the file has a generator's watermark. Just remember the flip side: the absence of a watermark proves nothing, so it's best to state explicitly in your brief whether AI is allowed or not.
5. Don't Argue in the Comments with Proof You Don't Have
A screenshot showing 'watermark not found' doesn't prove your video is real—and conversely, a detected watermark doesn't prove the entire scene was generated. Understanding this distinction is more useful than any screenshot.
What's Still Unknown
- When the detector's interface will be available in other languages; for now, it's English only.
- When exactly Apple Intelligence will be integrated; the company has said 'soon,' but no date is set.
- Whether other major video generators will join the SynthID initiative, as coverage remains partial without them.
- Whether platforms will begin reading SynthID directly upon upload to apply labels automatically.
- The official daily limit for free checks, which Google has not published.
Frequently Asked Questions
How can I check if a photo was made by AI?
Upload the file to synthid.com and see if an invisible SynthID watermark is detected. This works for images, videos, and audio created by AI from Google, OpenAI, NVIDIA, and Kakao. It won't work for generators from other companies, as the detector doesn't recognize their content.
Is it free and do I need an account?
The detector itself is free, but you need a Google, OpenAI, or Apple account to log in. The number of daily checks is limited—according to FourWeekMBA, it's around ten.
Can I remove the SynthID watermark from a video?
Not with standard editing—it survives cropping, filters, frame rate changes, and compression. And there's little point: the watermark itself isn't a violation, but failing to disclose AI-generated content on platforms is.
If the detector finds nothing, does that mean the video is real?
No. This is the most common mistake. The absence of a watermark only means one thing: the file does not contain a SynthID mark. The content could have been made by a generator that doesn't use this watermarking system.
Does the SynthID Detector identify AI text?
No, the public site works with images, videos, and audio. Google does not offer AI text detection here.
Further Reading
- AI for Video Editing: 15 Best AI Editors of 2026
- Free AI for Video Creation: GigaChat Makes Videos with Sound in Full HD
- How to Make Clips for YouTube Shorts: YouTube Is Slashing Reach for Re-uploads
AI transparency has become a professional requirement, not a matter of taste. On Prime Oracles, creators take on brand briefs in the 'Content Rewards' section and get paid per view (CPV). The brand specifies its rules on AI use in the brief itself, and it's worth reading them before you start editing.