Does ChatGPT add SynthID to images? How to check
ChatGPT images carry Google's SynthID watermark and OpenAI's Content Credentials. What each mark does, how to check a file for both, and what you can remove.
3 min read
On this page6 sections
Short answer. Yes. In 2026 OpenAI began applying Google DeepMind's SynthID watermark to images generated in ChatGPT, and Google's SynthID Detector was updated to read them. ChatGPT images have also carried C2PA Content Credentials signed by OpenAI since 2024. The credentials are metadata and can be removed; SynthID is in the pixels and is designed to stay. The on-device AI label checker shows the metadata; Google's detector shows the watermark.
What a ChatGPT image carries
| Mark | Added since | Where | Who reads it | Removable |
|---|---|---|---|---|
| C2PA Content Credentials signed by OpenAI | 2024 | The file's metadata | Instagram, Facebook, Threads, TikTok, LinkedIn, Google | Yes |
| IPTC digital source type (AI generated) | 2024 | XMP metadata | Meta | Yes |
| SynthID | 2026 | The pixels | Google's SynthID Detector | Not reliably |
| Visible watermark | Varies by plan | The picture | Viewers | By editing |
The first two are why an Instagram post of a ChatGPT image gets the "AI info" label. The third is new and is the reason "OpenAI SynthID" searches took off.
Why OpenAI adopted SynthID
Content Credentials are easy to lose. A screenshot has none, many editors drop them, and OpenAI said early on that metadata was not a complete answer. SynthID solves the other half: it is embedded in the pixels and survives the edits that strip metadata. Google opened SynthID to partners in 2025, and OpenAI's adoption means one detector now covers the two most used image generators.
For creators the practical change is this: removing the metadata from a ChatGPT image used to leave nothing for anyone to find. Now the pixels still carry a mark that Google can read.
How to check a ChatGPT image
For the metadata
Drop the file into the AI label checker. It runs in your browser and reports the Content Credentials (signed by OpenAI), the AI source type, any generator names and whether the file has camera data. This is what platforms read when they label a post. The Content Credentials Verify tool shows the full manifest if you want the signature details.
For SynthID
Upload the image to Google's SynthID Detector. It answers whether a SynthID mark is present and highlights the marked areas. OpenAI has also offered its own verification route for its images; use whichever is current. No metadata tool, including the checker, can see SynthID, because it is not stored in the file's metadata.
What you can remove, and what you cannot
The metadata. The image remover rebuilds the file, so the OpenAI manifest and the IPTC field are gone, camera metadata from a real iPhone is written in, and the picture gets an invisible variation that gives it a new fingerprint. After that, Instagram and TikTok have nothing in the file to label. First image free, up to 20 per batch.
SynthID. No published method removes it while leaving the image intact, and the sites selling "OpenAI SynthID removers" do not show before-and-after detector results. HideSlop does not claim to remove it. The detail is in can you remove SynthID?
Does it matter for posting?
Today's automatic AI labels on Instagram, Facebook, Threads and TikTok come from metadata, and none of the platforms has said it checks uploads for SynthID. So a clean file posts without a label. Google can detect the watermark, Google Search can surface it in "About this image," and platforms could start reading it. Treat SynthID as permanent and plan accordingly: if the content is realistic, the platform's disclosure rules apply to you whatever the file says.
FAQ
Does ChatGPT have SynthID?
Yes, for images generated since OpenAI adopted it in 2026. Older ChatGPT images carry Content Credentials but no SynthID.
Is there an OpenAI SynthID remover?
Nothing has been shown to remove SynthID while keeping the image. The metadata layer, which is what platforms read, can be removed.
Does a screenshot of a ChatGPT image remove the watermark?
It removes the Content Credentials. SynthID is designed to survive screenshots.
Does Google's detector work on ChatGPT images?
Yes. Google updated the SynthID Detector to read marks from partner models, including OpenAI's.
Will Instagram label a ChatGPT image?
Yes, from the Content Credentials and the IPTC AI field. With the metadata removed, there is no automatic label. Realistic content should still be disclosed under Meta's rules.