Do ChatGPT Images Have SynthID? A SynthID Detector Guide

Yes. Images from ChatGPT, Codex and the OpenAI API carry a SynthID watermark and Content Credentials. How to check one and what each result means.

Oct 21, 2026

Yes. Images made with ChatGPT, Codex and the OpenAI API carry Google DeepMind's SynthID watermark, and they carry Content Credentials (C2PA) as well. Google's official SynthID Detector recognizes them, and a metadata reader can often name OpenAI as the source while the credentials are still attached.

Why ChatGPT images carry SynthID

SynthID began as Google's watermark for its own models. OpenAI now adds it to its images, and NVIDIA (for its Cosmos video models) and Kakao have adopted it too. For anyone checking a picture, the practical effect is that one detector covers images from both Gemini and ChatGPT.

Two signals in every ChatGPT image

The SynthID watermark sits inside the pixels. It says a SynthID-enabled model made or edited the image, and it is designed to survive compression, resizing, cropping and filters. Many stacked edits can still weaken it. Only Google's tools can read it, because Google has not released a public detection API.
Content Credentials are signed metadata stored in the file. They name the company that signed them, here OpenAI. Screenshots and most social uploads strip them.

The two fail in opposite ways, so checking both tells you more than either one alone.

How to check a ChatGPT image with the SynthID Detector

1. Read the Content Credentials

Drop the original file into the metadata checker on our homepage. It reads the file in your browser without uploading it. If the Content Credentials are intact, the report shows OpenAI as the signer and may also list a GPT Image or DALL·E generator tag.

2. Confirm the watermark on synthid.com

Go to synthid.com and sign in. If you use ChatGPT, your OpenAI account works there, as do Google and Apple accounts. Upload the image (JPG, PNG, WebP, HEIC, AVIF, TIFF, GIF or BMP, up to 100 MB) and wait for the "detected" or "not detected" answer. The service is free, with a daily limit on checks.

The Gemini app has a built-in check too, but it covers Google AI content only, so use synthid.com for ChatGPT images.

3. Compare the two answers

When both agree, you have strong evidence. When they disagree, the pattern usually tells you what happened to the file.

What each SynthID Detector result means

Watermark detected, OpenAI credentials present. The clearest case: an OpenAI model made or edited the image, and the file has not been stripped.
Watermark detected, no credentials. Common for images that were screenshotted or posted to social media. The watermark survived, but the signed record that names OpenAI did not.
No watermark, OpenAI credentials present. The credentials still name OpenAI, which is evidence in itself. The watermark may have been weakened by heavy edits, or the image may predate OpenAI's use of SynthID.
Neither signal. Not proof that the image is human-made. It may come from a model that does not use SynthID, such as Midjourney, Stable Diffusion, FLUX or Grok, or it may have been altered.

Google also notes that in very rare cases SynthID can trigger on content without a watermark, so treat any single result as one piece of evidence. Our guide on how to read a SynthID Detector result goes into more detail.

What about older ChatGPT images and Sora?

An image made before OpenAI started adding SynthID will not carry the watermark, so "not detected" on an old ChatGPT image says little. The SynthID coverage Google lists for OpenAI names images from ChatGPT, Codex and the API; this guide makes no claim about Sora video. For AI video from Google, see how to check a Veo video.