SynthID Detector

The SynthID Detector is Google's public verification portal, at synthid.com. Upload an image, a video, or an audio file, and it tells you whether that file carries a SynthID watermark — the invisible mark that Google's AI tools, and a growing list of partners including OpenAI, embed in the content they generate.

Google opened the portal to everyone on October 7, 2026, in English worldwide. Before that, it spent 2025 as a limited trial for journalists and media professionals. The scale behind it is worth keeping in mind: Google says it has watermarked more than 180 billion images and videos and 240,000 years of audio to date. That is the pool this tool can speak about — which is also why its answer is narrow, and why that narrowness is the point.

What SynthID is, and who embeds it

SynthID is Google DeepMind's watermarking system, launched in 2023. When a supported model generates content, it weaves a signal directly into the output — into the pixels of images and video, and into the frequency components of audio. The mark does not change how the content looks or sounds, and it is designed to survive the normal life of a file online: cropping, filters, color changes, and lossy compression such as JPEG.

That is what separates it from metadata. Provenance metadata like C2PA Content Credentials can be stripped by a screenshot or a re-export. A watermark woven into the pixels travels with the content itself.

Who embeds it today:

On October 15, 2026, TIME named SynthID one of its Best Inventions of 2026.

How to use the SynthID Detector

  1. Go to synthid.com. That is the portal itself — a website, not an app.
  2. Sign in. The portal accepts Google, OpenAI, and Apple accounts. Checks are tied to that account.
  3. Upload your file. Images, video, and audio are all accepted. The full format list is in the next section.
  4. Read the verdict. You get one of two answers: a watermark was found, or it was not. The public site gives you only that yes-or-no; Google's internal version of the tool highlights which part of a file carries the watermark, and that highlight is not part of what you get here.

Two practical notes. Use the highest-quality copy you have, because the portal warns that heavy compression and heavy editing degrade the signal it reads. And on privacy, the terms are specific: uploaded media "are processed in real-time," and once results are returned "the original media files are immediately deleted." A digital signature of your file is kept for 24 hours to enforce quotas, then permanently deleted.

What files the portal accepts

This is the first thing most people ask, and most pages about the tool don't answer it. The portal's supported formats, as listed on its upload panel:

HEIC and HEIF are on the list — the default format for iPhone photos. You don't need to convert an iPhone photo before checking it, and you don't need to re-export a video or transcode audio first.

Three other ways Google lets you check

The portal is not the only door. Google describes watermark checks built into three of its surfaces, which together handle over one million requests a day:

The portal is the option that doesn't require a conversation with an AI model, which matters if you don't want to take an AI's word for its own work.

What your result actually means

Watermark detected. The file was generated or edited by a SynthID-enabled model. This is strong evidence: the signal is embedded at creation time, so a positive means a participating AI tool was in the file's history.

Watermark not detected. This is where people misread the tool. A negative result does not mean a human made the file. The portal lists the reasons a genuinely AI-generated file can still come back negative:

In other words: detected is evidence, not detected is the absence of evidence. Treat a negative as "unverified," not as "real."

Is the SynthID Detector accurate?

For what it actually checks, yes — and for a structural reason. Regular AI detectors analyze patterns and artifacts in the pixels to estimate how likely content is AI-made, which is where false positives come from: a real photo can score as AI. The SynthID Detector doesn't estimate anything. It reports whether a specific, embedded signal is present, so a real photo won't trip it.

The trade-off sits on the other side of the ledger. Because it only recognizes watermarks from participating generators, it says nothing about AI images from tools that don't participate. OpenAI's own verification page, which works the same way, makes the same point: detection is highly reliable for the signals it covers, and silent about everything else.

Practical accuracy tips, from the portals' own guidance: don't crop the image or convert it to another format before uploading, upload one file at a time, and for audio, OpenAI recommends clips between 10 and 60 seconds.

Does ChatGPT use SynthID?

For images and audio, yes. Google's announcement names OpenAI as a partner, and OpenAI's verification page confirms that its tools check for SynthID watermarks and C2PA Content Credentials in content generated with ChatGPT, the OpenAI API, and Codex — images and audio files.

For text, no — not through this route. OpenAI's verification page states that OpenAI.com/verify does not currently support text verification, and that organizations with a qualifying use case can apply for access to text verification through the Content Provenance API. SynthID text watermarking does exist on Google's side, applied to output from the Gemini app and web experience, where it works by adjusting the word-probability choices the model makes as it writes.

SynthID Detector vs. regular AI image detectors

They answer different questions with different mechanisms:

Use the watermark check first when you have it, use pattern-based tools for the cases it can't speak to, and weight the outputs accordingly.

Is this the same thing as "Google's AI image detector"?

If you searched for a Google AI image detector, this is what exists in 2026. Google does not ship a standalone general-purpose "AI image detector." Its public answer to that need is SynthID — the watermark checks in Search, Chrome, and Gemini, plus this portal. If you need to flag AI images from tools that don't watermark, you need a third-party detector, which is a different product category.

Limits to know before you trust a result

Is it free?

Yes. The portal is free to use with a signed-in account. Google hasn't published a number in the terms — they only refer to "daily usage quotas." Google told Ars Technica that each user gets a daily usage quota of approximately 10 image video and audio checks. Treat the terms as binding and the figure as what Google says in practice, since it can change without a terms update.

FAQ

How do I detect SynthID? Four ways: the portal at synthid.com, asking Gemini in a chat with the file attached, and the verification information built into Google Search and Chrome. The portal is the one built for the job.

How can I check if an image has SynthID? Go to synthid.com, sign in, upload the image, read the verdict. Use the best-quality copy and don't crop or convert it first.

Does it accept HEIC photos from an iPhone? Yes. HEIC and HEIF are both on the supported list, alongside JPG, PNG, WEBP, AVIF, TIFF, BMP and GIF. No conversion needed.

Does it work on Midjourney or Stable Diffusion images? No. Those generators don't embed SynthID, so the watermark check has nothing to read. You'd need a pattern-based detector, and should treat its output as a probability, not proof.

Is there an app or an API? The portal is web-based and Google hasn't announced a public API for it; its terms explicitly prohibit automated and bulk use. OpenAI, separately, does offer programmatic verification through its API — for OpenAI-generated content.

Does the watermark survive a screenshot? SynthID is designed to survive cropping, filters, and lossy compression. Third-party tools that read the watermark report it also survives screenshots, since a screenshot re-encodes pixels rather than stripping a metadata tag — but heavy modification can still degrade the signal.

Sources

  1. Google blog, October 7, 2026 — Google expands SynthID Detector for AI content: blog.google/innovation-and-ai/models-and-research/google-deepmind/synth-id-ai-content/
  2. Google blog, October 15, 2026 — TIME has named Google SynthID one of the Best Inventions of 2026: blog.google/innovation-and-ai/models-and-research/google-deepmind/synthid-time-best-inventions/
  3. Google DeepMind — Google SynthID overview: deepmind.google/models/synthid/
  4. SynthID Detector portal, including supported formats and additional terms: synthid.com
  5. OpenAI — Verify OpenAI-generated content: openai.com/research/verify/
  6. Engadget — How to use Google's SynthID AI detection website
  7. Ars Technica — reporting on the SynthID Detector global rollout, October 2026

Links on this site may be affiliate links.