Google has opened its SynthID Detector to everyone. Since 7 Oct 2026, anyone can go to synthid.com and check whether an image, video or audio file was made or edited with AI from Google or its partners, which Google names as OpenAI, NVIDIA, Kakao and, "soon", Apple.
Start with the limit, because it matters more than the launch. This is not a general AI detector. Google's own site says so in those words. It only spots SynthID watermarks from companies that use the technology, and a "not detected" result doesn't prove a person made something. It doesn't check text either.
For freelancers who handle client media, the call is simple. Check client-supplied or stock files when provenance matters. Don't rely on a clean result as proof of human work. Disclose your own AI use, because clients can now check in seconds. Skip it if you only ship your own raw originals, and wait if you need Apple media or text covered.
Affiliate disclosure: some tool or platform links may later be affiliate or referral links. Facts in this post come from Google's launch blog, synthid.com and Google DeepMind's SynthID page, with secondary reporting where marked, checked on 8 Oct 2026 (SAST). Not legal advice.
What changed on 7 Oct
An earlier version of the Detector launched last year for journalists and media professionals. Now, in the words of Google DeepMind's Pushmeet Kohli on the Google blog, "we're expanding access to everyone – with the tool available globally in English starting today."
The big shift is the partner list. Google says anyone can check whether a file "was made with AI from Google or our partners, including OpenAI, NVIDIA, Kakao, and soon, Apple." Google gives no date for Apple. It also names companies, not products, so don't assume every image from a given partner app carries a watermark.
For scale, Google says it has watermarked over 180 billion images and videos and 240,000 years of audio since 2023. TechCrunch lists Google tools that add SynthID, including Nano Banana, Veo, Lyria, Gemini, Flow and Vids. Our Gemini 3.8 Flash TTS guide for freelancers already notes that Gemini Audio clips carry SynthID watermarking.
What it can and can't tell you
The Detector handles three content types. For images, it looks for the presence of a watermark. For video and audio, it looks for segments that carry one. On the upload screen, video is labelled Preview, so treat video results as early.
Here's how synthid.com explains the two results:
- Watermark detected: a SynthID watermark is present, and the content is "likely edited or generated by a SynthID-enabled AI model".
- Watermark not detected: no confident signal. The site says the file is "not likely to have been edited or generated by a SynthID-enabled AI model", then adds: "this does not totally rule out AI generation or manipulation."
Why might AI-made media come back as not detected? The site lists the reasons: it was made by a model outside the partner group, made before a company adopted SynthID, or edited so heavily that the watermark degraded. And it works the other way too. "In very rare cases SynthID can also falsely trigger on non-watermarked content." Google gives no accuracy or false-positive figures.
What it won't do:
- Check text. SynthID can watermark Gemini text, but the Detector only accepts images, video and audio. It won't tell you whether a blog post, essay or proposal was AI-written.
- Catch every generator. Google's site is blunt: "We can only detect media from the above companies, who have adopted SynthID technology."
- Settle a dispute alone. Google calls SynthID "one important layer" and "not a standalone guarantee", and recommends "collecting multiple points of data before making a decision."
On robustness, Google says the watermark is designed to survive filters, colour changes and lossy compression, plus cropping and frame-rate changes for images and video, and noise, MP3 compression and speed changes for audio. But "if someone tries many transformations, they may be able to find one that doesn't get detected."
How to run a check properly
- Sign in. synthid.com asks you to sign in before detection, with a Google, Apple or ChatGPT account.
- Get the best file you can. The site says to use "the highest quality file possible", because heavy compression and editing affect detection. Ask for the original export, not a screenshot or a copy forwarded through a chat app.
- Check the format. When we looked on 8 Oct, the upload screen listed a 100 MB maximum per file. Images: JPEG, PNG, BMP and WebP. Audio: MP3, WAV, OGG, AAC, M4A and FLAC. Video: MP4, MOV and WebM, marked Preview. TechCrunch lists more image formats, so the site may accept others, but HEIC (the iPhone default) isn't on the site's own list. Check the live upload screen before you assume.
- Mind the quota. A daily quota applies. Ars Technica reports Google said users get roughly 10 image, video and audio checks a day. Google's own pages don't give the number, and the site's error messages suggest you'll be asked to wait some hours once you hit the limit.
- Think before you upload client work. We found no detector-specific note on how uploads are stored. The site links Google's Terms and Privacy Policy. Don't put confidential client files through it until you've read those and you're comfortable.
On cost: Google's blog and synthid.com show no price or paid tier at launch, and Engadget reports the tool is free to use. We wouldn't promise it stays that way.
Where it fits a freelancer workflow
At intake. A client sends a "real photo" for a campaign, a testimonial video or a voice clip. A quick check before you build on it costs a minute. If it comes back detected, ask the client about it before you go further. Don't accuse anyone; ask.
Before delivery. If a project is news-adjacent, makes ad claims, or the client has said "no AI", run your sourced assets through it as one QA step. Note the result in your job folder.
On your own AI work. If you used Google tools or partner tools to make or edit something, assume your client can check it in seconds. Put AI use in the brief or statement of work up front. We made the same point about disclosure in our Firefly AI Assistant try-or-skip guide, and it applies even more now.
Keep the other evidence. A Detector result is one data point. Source files, edit history, C2PA or EXIF data where present, and a written client warranty about where material came from all count for more together than any single check. None of this is legal proof of authorship.
Other ways to check
Google has also built SynthID verification into Search, the Gemini app and Chrome. You upload a file and ask whether it was made or edited by Google AI. Google says these surfaces handle over 1 million requests a day. They're handy for a quick look at Google-made media when you've used up your daily checks, but remember they're asking about Google AI specifically.
Check, don't rely, disclose, skip or wait?
Check client-supplied or stock media before you ship it, whenever provenance matters. Use the highest-quality original. Detected means talk to the client.
Don't rely on "not detected" as proof of human work. Generators outside the partner list won't show up; Midjourney, for example, isn't one of the partners Google names. If you're comparing image tools, our Midjourney vs free alternatives roundup covers that side. Heavy edits and compression can also strip the signal.
Disclose your own use of Google or partner AI tools. A client check takes seconds, so the SOW is a better place for that conversation than an awkward email later.
Skip it if you only shoot and record your own originals and keep the raw files, or if you need bulk checks. With a daily quota, it's a spot-check tool, not a pipeline.
Wait if your worry is Apple-generated media (support is "soon", no date) or AI-written text, which the Detector doesn't cover.
| Who you are | Check | Don't rely on | Disclose | Skip or wait |
|---|---|---|---|---|
| Designer using stock packs | Hero images and anything sold as "real" | A clean result on non-partner stock | Any AI edits or generated fills | Skip for your own shoots with raw files |
| Video editor with client footage | Key clips, knowing video is Preview | Re-encoded or heavily edited exports | AI-generated b-roll or fixes | Wait if you need more than a spot check |
| Podcast editor with voice clips | Guest or client-supplied audio segments | Low-bitrate copies from chat apps | AI voices, cleanup or generated music | Skip for raw recordings you captured |
| Social manager with UGC | Customer photos used in ads or testimonials | Screenshots and reposted copies | AI-made graphics in the content plan | Wait for Apple media support |
The Joburg reality check
Google says the Detector is available globally in English. It hasn't published any country exclusions, so it should work from South Africa like anywhere else, with the same sign-in.
The practical issue here is files, not access. A lot of client material reaches South African freelancers through WhatsApp, which compresses photos, video and voice notes. Since compression hurts detection, ask for the original file or a cloud link instead. Then keep a short provenance note in the job folder: where the file came from, when you checked it, and what the result was. That's how we'd run it from a Joburg desk. It works the same anywhere.
FAQ
Is SynthID Detector free?
Google's pages show no price or paid tier at launch, and Engadget reports it's free to use. That could change, and a daily quota applies.
Does "not detected" mean a human made it?
No. It means no SynthID watermark was confidently found. The media could come from a non-partner tool, predate SynthID adoption, or have been edited enough to weaken the watermark.
Can it check ChatGPT or Midjourney images?
It can only detect media from SynthID partners, and Google names OpenAI as one. Google doesn't list which OpenAI products carry the watermark. Midjourney isn't on Google's named partner list.
How many checks do I get?
A daily quota applies. Ars Technica reports Google said roughly 10 image, video and audio checks a day. Google's own pages don't state the number.
Sources: Google expands SynthID Detector for AI content (Google Blog, 7 Oct 2026), SynthID Detector and FAQ (synthid.com), SynthID (Google DeepMind), Google rolls out improved SynthID AI content detector (Ars Technica, 7 Oct 2026), Google's SynthID Detector website is now available (Engadget, 7 Oct 2026), Google's new SynthID website can identify AI-generated media (TechCrunch, 7 Oct 2026). Facts checked 8 Oct 2026 (SAST); re-check the upload screen, formats and quota wording at paste time.