OpenAI and Gemini Image Verifiers: What SynthID Checks Miss

Until October 2026, the advice for checking an AI image was to pick the verifier that matched the generator: OpenAI Verify for ChatGPT images, Google's tools for Gemini images. That advice made sense when each provider's checker read its own signals. On October 7, 2026, Google changed the picture.

Google announced that its SynthID Detector now lets anyone check whether an image, video, or audio file was made with AI "from Google or our partners," and it names OpenAI, NVIDIA, Kakao, and, soon, Apple among those partners. You can reach it at synthid.com, and verification is built into Search, the Gemini app, and Chrome. The launch is global, in English only.

So one SynthID check can now cover both OpenAI and Gemini images. The harder question remains: what happens when the file carries no SynthID mark at all? This guide covers what OpenAI Verify reads, what Google's SynthID Detector now covers, the gap a SynthID-only check leaves, and how to close it. Want to try a file from a browser first? Use the free online AI image detector.

What OpenAI Verify Actually Tells You

OpenAI Verify checks supported C2PA metadata and SynthID watermarks associated with OpenAI tools. A positive result answers a real question: did this file likely come from OpenAI? OpenAI says its generated images include C2PA metadata and SynthID watermarks, which is why that answer can be specific.

A negative result is narrower than it looks. It only means the file does not carry the OpenAI signal the tool can read, so Gemini, Adobe Firefly, Midjourney, Stable Diffusion, or another generator could still be involved.

For the step-by-step OpenAI workflow, read how to detect ChatGPT-generated images.

What Google's SynthID Detector Now Covers

Google's SynthID Detector reads SynthID marks from Google and from partners that stamp their media with SynthID, OpenAI included.

SynthID is an invisible watermark embedded directly into generated media. Metadata travels beside the pixels; SynthID lives inside them. That is why a SynthID mark can survive a screenshot or a re-encode that would wipe a metadata record. It is still not magic: heavy edits, severe recompression, and tiny crops can push a result to inconclusive.

Google's announcement covers who stamps the mark and where to check it, so read a result for exactly what it says.

What a SynthID Check Still Misses

A SynthID check can only find images that were stamped with SynthID. A missing mark proves nothing about who or what made the file.

That leaves three gaps, and each one shows up in ordinary use. The first is unstamped generators. Stable Diffusion running locally in tools such as ComfyUI or Draw Things adds no SynthID mark, so a SynthID check has nothing to read. The second is edited files. A SynthID mark can degrade under heavy editing, and a file that went through several apps may come back inconclusive. The third is the file with no story: a screenshot from a group chat, a marketplace product photo, or a CMS export with no source label. A blank SynthID result on that file tells you only that no SynthID mark was found.

Anthropic is a useful example of why one signal is never the whole picture. Supported Claude image files carry C2PA Content Credentials, not image SynthID. Our Claude watermark guide explains what each signal proves.

Why C2PA and IPTC Metadata Can Disappear

C2PA Content Credentials are a signed credential attached to the file, which makes them a different kind of evidence from a watermark. A credential can name the tool that created or edited an image and carry more context than a watermark can. IPTC metadata can also preserve useful file details.

The weak point is ordinary file handling. Screenshots, social apps, exports, and image optimizers often strip metadata. A credential that survives is strong evidence once its signature validates against a trusted signer. A credential that was stripped leaves no trace, which is why a metadata check and a SynthID check answer different questions. To run both in one pass, see how to check an image for an AI watermark.

How Slop or Not Checks AI Images

Slop or Not checks every layer of evidence before it guesses, so generators that stamp nothing still get a result:

1. It reads SynthID from Google and OpenAI images.
2. It verifies signed C2PA Content Credentials against a bundled trust list, so a forged credential fails.
3. It reads embedded generator fingerprints from the file metadata, which names tools like Stable Diffusion WebUI, ComfyUI, and Draw Things.
4. It scans the pixels with its AI image detection model when provenance is missing or unclear.

The pixel layer returns a probability, not proof. In our internal tests it reached 90% accuracy on images without a watermark; results can vary with new AI models and methods designed to trick detectors.

Already suspect a provider? Use the ChatGPT image detector for suspected OpenAI images or the Gemini image detector for Google generation. Use the SynthID detector when the watermark itself is the question, and the AI watermark detector when you want watermark and metadata evidence separated before the pixels are judged. If the source is unknown, start with the broad AI image detector.

Online Checker or Native App?

Use the free online image checker when you want a quick browser result. The check runs on Numen's private Mac server and is never passed to OpenAI, Google, or another AI company.

Use the Slop or Not app when the photo should stay private. The iPhone, iPad, and Mac apps run every layer on-device, so nothing leaves the device. That is the better path for student work, unpublished reporting, client files, private messages, and personal photos.

FAQ

Can OpenAI Verify detect Gemini images?

No. OpenAI's verifier is built for supported OpenAI provenance signals. A missing OpenAI signal only tells you that OpenAI Verify did not find what it can read.

Can Gemini detect OpenAI SynthID?

Yes, as of October 7, 2026. Google's SynthID Detector, built into the Gemini app, Search, and Chrome and available at synthid.com, now checks for SynthID from Google and its partners, OpenAI included. It can only find images stamped with SynthID, so a clean result does not clear an image.

Is SynthID the same as C2PA?

No. SynthID is an invisible watermark embedded into generated media. C2PA Content Credentials are signed metadata records. They work well together, but they are different kinds of evidence.

What if no watermark or metadata is found?

A missing signal is not proof that a person made the image. Metadata may have been stripped, a watermark may have degraded, or the image may come from a generator that stamps nothing. Slop or Not then reads generator fingerprints and scans the pixels, which returns a probability rather than proof.

Which Slop or Not checker should I use first?

Use the broad AI image detector when the source is unknown. Use the ChatGPT image detector, Gemini image detector, SynthID detector, or AI watermark detector when you already know the narrow question.

Check the Image Without Guessing the Generator

Google's SynthID Detector is a real step forward, and it answers one question well. Start with the free online AI image detector for a browser check, or download Slop or Not for iPhone, iPad, and Mac to check every layer on-device.