What Google puts into Gemini images
Google embeds SynthID, its imperceptible watermark, into content from its generative models. In May 2026 it said SynthID had been applied to over 100 billion images and videos and 60,000 years of audio. Google also uses C2PA Content Credentials across a growing number of its generative media tools; these signed records state how the media was created and, in Google's images, that a SynthID watermark was applied.
The watermark is inside the pixels and is designed to survive edits such as compression or screenshots. The Content Credentials sit next to the pixels and are removed by most screenshots and social platforms, but while present they are signed and name the signer.
How to check a Gemini image
Google's help page for Gemini lists what you need and how it works. You must be signed in, using a personal or qualified Workspace account and the latest app version.
- Open Gemini Go to gemini.google.com or open the Gemini app.
- Add one file Upload a single image, video or audio file of 100 MB or less. Videos must be under 90 seconds and audio under one hour. If you use a screenshot, crop tightly around the image and do not upload a collage.
- Ask the question Ask, for example, 'Was this image created or edited by Google AI?' or 'Is this AI-generated?'.
- Read the answer Gemini reports whether a SynthID watermark was detected, and for videos, in which parts. Content Credentials verification is available on gemini.google.com and in the Android app, with iOS to follow, according to the help page.
What OpenAI puts into ChatGPT images
OpenAI's help center says images generated with ChatGPT, Codex or the OpenAI API include both C2PA Content Credentials and SynthID watermarks, using Google DeepMind's SynthID technology. OpenAI explains the reasoning: C2PA carries detailed context, while the watermark helps preserve a signal when metadata does not survive. OpenAI also notes that coverage varies by product, model, export path, file type and creation date, and that API outputs from models planned for imminent deprecation, including gpt-image-1 and gpt-image-1.5, are not covered.
How to check a ChatGPT image
Upload the image at openai.com/verify. OpenAI says the tool looks for a SynthID watermark that originates from OpenAI or a trusted C2PA manifest from OpenAI, and that if it finds either, the image was generated by OpenAI tools. It confirms only that origin; it does not say whether the image is accurate, unedited or used in the right context.
Developers can run the same check with OpenAI's Content Provenance API (POST /v1/content_provenance_checks). Its C2PA result reports a validation state of trusted, valid, invalid or not_present, and the outcome is detected only when a trusted or valid manifest names OpenAI as issuer and includes an AI-generation action.
Using the SynthID Detector on this site
This independent site reads the parts both companies leave in the file: it validates C2PA Content Credentials against the official C2PA trust list, shows the signer (for example Google LLC or OpenAI), shows whether the credentials declare a SynthID watermark, and scans EXIF, XMP and PNG metadata. That check is free. It does not decode the pixel watermark itself, so if the credentials have been stripped, use the Gemini app or openai.com/verify on the same file.
Reading the results
A detected watermark or a trusted manifest from Google or OpenAI is strong evidence that the image came from that company's AI. A negative result is narrow. Gemini's help page says a missing SynthID watermark means the content was not created or edited by Google AI but could come from other AI systems, and lists unclear cases such as very simple content or edits that are too small. OpenAI's documentation says a not-detected result does not rule out OpenAI content whose metadata was stripped or whose watermark degraded, and does not cover other companies' models.