IMAGE PROVENANCE GUIDE
SynthID Detector: How to Use It, Partners & Limits
Google’s SynthID Detector opened to everyone in October 2026. Learn where to find it, sign-in and daily limits, whether it detects ChatGPT (OpenAI) images, and how SynthID watermarks differ from C2PA markers.
How to check a SynthID watermark
- Start from the original file rather than a screenshot or a social-media repost.
- Open synthid.com, sign in and upload the image, video or audio. Alternatively, upload it in Gemini and ask whether it was created or edited with Google AI.
- Read the exact wording of the result: watermark detected, not detected, or inconclusive.
- Check the same file with AIGCTrace for C2PA, generator names and readable metadata, which covers sources SynthID cannot.
- Save the tool used, check time, file version and full result. Keep watermark findings, credential information and visual observations separate.
What changed in October 2026?
According to Google’s SynthID page and press coverage, SynthID Detector opened to everyone on October 7, 2026, after testing with journalists and media professionals. It checks images, video and audio. Press reports say it now requires signing in and allows about 10 checks per person per day. Check synthid.com for the current sign-in options, limits and languages. Google also offers verification in Gemini, Google Search and Chrome, and text watermarking now covers the Gemini app and web.
What does a SynthID detector check?
SynthID is Google DeepMind’s invisible watermarking technology. The watermark is embedded at generation time and is designed to survive common edits such as cropping, filters and lossy compression. SynthID Detector looks for that watermark and answers whether content was made or edited with AI from Google or its partners. It cannot recognise providers that do not use SynthID, and it does not establish that a depicted event happened.
Which providers can it detect?
Google DeepMind lists OpenAI, NVIDIA and Kakao as SynthID partners, with Apple coming soon. The official Detector can recognise SynthID-marked content from those partners. Midjourney, Stable Diffusion, Flux and other systems that do not use SynthID will not be flagged, even when their output is AI-generated. Gemini’s verification help focuses on Google AI tools, so confirm each entry point’s coverage separately.
SynthID watermarks versus C2PA metadata
A SynthID watermark is embedded in the content signal. C2PA credentials carry provenance claims with a separate signature validation process. Images from Google’s Nano Banana Pro and Nano Banana 2 carry both SynthID and C2PA; OpenAI images carry C2PA and have reportedly adopted SynthID. C2PA metadata is easily lost in screenshots and reposts, while the watermark is designed to survive processing. A SynthID or c2pa text match is neither a detected watermark nor a validated credential. Record the two checks separately.
Why keep the original file?
An original allows you to compare watermark results and readable file markers. A screenshot creates a new file and may lose metadata; watermark detection depends on the processing applied. Missing metadata does not establish that a SynthID watermark was removed. Preserve each version and record which one was checked.
What can AIGCTrace add?
Inspect C2PA markers, generator names from Google (Gemini, Nano Banana, Imagen), OpenAI and others, China’s GB/T 45438-2025 AIGC labels and readable metadata, alongside pixel heuristics, screenshot features and statistical signals. When a Google or OpenAI source is identified, the report suggests cross-checking with the official SynthID Detector. The online tool accepts PNG, JPEG, WebP, GIF and TIFF up to 15 MB. Pixel analysis requires decodable PNG/JPEG with a long edge at most 3000 pixels. These checks do not decode SynthID or validate C2PA signatures.
Choose a check for your question
| Method | Question it addresses | Scope |
|---|---|---|
| Official SynthID Detector | Does the file carry a SynthID watermark from Google or a partner? | Sign-in and daily limits apply; covers only providers that use SynthID. |
| Gemini / Search / Chrome verification | Is there a Google AI SynthID provenance signal? | Confirm its coverage separately from the standalone Detector. |
| C2PA credential validator | Do the credentials and signatures validate? | Valid credentials do not establish the truth of the depicted event. |
| AIGCTrace image analysis | What readable markers and heuristic clues are present? | Covers Midjourney, Flux, China AIGC labels and more; no SynthID decoding or C2PA signature validation; scores are not AI probabilities. |
SynthID Detector FAQ
Can SynthID Detector detect ChatGPT images?
You can try. OpenAI is a SynthID partner listed by Google, press reports say it has adopted SynthID, and the official Detector can recognise SynthID-marked OpenAI content. Older files may not carry the watermark, so rely on the Detector’s actual wording. ChatGPT images usually also carry C2PA credentials; AIGCTrace can show markers such as OpenAI and gpt-image.
Do I need to sign in? Is there a daily limit?
Press reports say the public Detector requires signing in and allows about 10 checks per person per day. Sign-in options and limits may change, so check synthid.com. For larger batches, screen provenance markers with AIGCTrace for free, then verify the key files with the official tool.
Does no SynthID watermark mean a real photo?
No. The file may come from an AI system that does not use SynthID, or processing may affect the signal. Record the negative result for the file checked; do not turn it into proof of camera capture.
Is AIGCTrace a free SynthID detector?
No. AIGCTrace offers free image provenance checks but does not decode SynthID pixel watermarks; a SynthID entry in its report only means the text appears in the file. Use Google’s official link above for SynthID verification and this site for additional readable markers and image clues.
Does a watermark mean the whole image was AI-generated?
Do not assume that. A file may include partial AI edits or combined content. Keep the tool’s precise wording and interpret it alongside the original, publication source and editing records.
Sources and scope
Product limits reflect the current project implementation. Illustrative examples are not accuracy benchmarks.