What is a SynthID detector and how does it work?
SynthID is Google DeepMind’s family of watermarking technology for AI-generated content. It embeds a pattern that marks content as machine-generated. The pattern sits either in the pixels of an image or alongside the file as a written declaration. Google applies SynthID to output from its own tools, including Gemini and Google’s AI image generators. Other companies attach related provenance declarations of their own.
A SynthID detector works at whichever layer the file still carries. A written declaration, such as a C2PA Content Credential, an IPTC label, or an embedded provenance note, is readable metadata. A checker parses the file and reports what the declaration states. The invisible pixel watermark is a different matter. Verifying the pixel watermark requires Google’s proprietary detector, which third-party tools cannot run. No independent SynthID watermark detector should claim to confirm it. A practical checker reads every layer it can reach and names the one it cannot.
This page runs both checks on your file and returns a single report. The signal report reads what the file declares about its origin. The metadata listing shows the layers underneath it.
How to detect SynthID in an image
Run the check on the original file, not on a screenshot or a re-shared copy. Every hop through a messaging app or a social platform can strip the declarations a checker reads. The most reliable test uses the image exactly as you received it from its source.
Upload the image and read the two layers the report returns. The signal report summarizes the file’s own account of its origin. It says whether provenance records are present, which platform or generator they name, and how strong the combined evidence is. The metadata listing shows the raw layers underneath, including C2PA Content Credentials, EXIF generator fields, IPTC labels, and any embedded watermark statements. The listing is the evidence itself. Nothing gets compressed into a single score.
Reading the result takes about a sentence. A file whose credentials name Google’s generator, for example, is declaring SynthID provenance, even though only Google can verify the pixel watermark itself. An empty report describes this particular file. Every re-shared copy counts as a separate file.
Which evidence layers the check reads, and which stay out of reach
C2PA Content Credentials
A third-party check reads a C2PA manifest like any other metadata and reports what it declares, including which tool made the file and whether the credential is intact.
SynthID declaration in metadata
A written SynthID declaration travels in readable metadata, so an independent check can parse it and report what it states.
SynthID pixel watermark
The file can declare the invisible pixel watermark, and only Google’s official portal can verify the pattern itself. No third-party tool should claim to confirm it.
OpenAI SynthID
Some OpenAI outputs declare a SynthID-style signal in readable metadata. The signal itself is OpenAI’s to verify. This checker reports the declaration and does not test the pixels.
Why an empty report cannot prove an image is real
A SynthID check reads evidence, and evidence can be missing. An AI image with its metadata stripped leaves nothing to read. Plenty of AI images circulate in that state: exported, compressed, screenshotted, or passed through platforms that discard provenance on upload. On those files, every declaration-based check returns the same answer: nothing was found.
The reverse holds too. A real photograph can carry no metadata at all. Cameras vary widely in what they embed, and some publishing workflows strip everything before release. An empty report proves nothing either way. It cannot confirm an image is real any more than it can confirm it is synthetic.
Treat the report as one input to a larger judgment. For high-stakes decisions, pair it with what you can learn about where the file came from. Use Google’s own tool when the question is about the watermark itself. No single signal, this one included, settles the question.
How it works
How to run a check and read the report
Step 1: Choose an image
This checker takes PNG, JPEG, WebP, BMP, TIFF, HEIC, HEIF, and AVIF images up to 20 MB.
Tip
Originals straight from the source preserve the most readable evidence.
Step 2: Run the check
Start the job, and both operations run at once. The signal report covers the file’s provenance declarations, and the forensic listing covers its metadata layers. A full check costs 10 credits, and the panel shows the exact price before it runs.
Tip
Credits are only consumed when a check actually runs.
Step 3: Read the report
Each check returns its own card. The AI-generation signal verdict comes with its supporting declarations, and the metadata listing shows which provenance records are present or missing. If the file declares a Google SynthID watermark, confirm it through Google’s official detector. That portal is the only channel that can.
Tip
Google’s check covers the pixel watermark alone and does not read C2PA manifests or EXIF history.
SynthID checker vs Google’s official detector
Only Google can rule on the pixel watermark. The Google SynthID detector runs on Google’s systems and stays closed to third parties. It answers one narrow question: does this file carry Google’s invisible SynthID watermark? Google’s detector reads that one watermark. C2PA manifests and other vendors’ provenance layers fall outside its answer.
This SynthID checker does the complementary job. As an independent third party, it reads the portable evidence that travels with the file: C2PA Content Credentials, EXIF and IPTC fields, and written SynthID or OpenAI declarations. Everything lands in one report, including missing evidence. A stripped file is a finding in itself.
Use both when the stakes are high. Start here for a fast, readable inventory of what the file itself declares. Go to Google’s tool when you need the watermark verdict that only Google can give.
What a declaration-based check cannot prove
FAQ
Frequently asked questions
Plain answers about what a SynthID detector can read and what only Google can confirm.
Explore more tools
SynthID Remover
Strip SynthID declarations and AI provenance from generated images.
AI Image Detector
Classify whether an image is AI-generated, with per-signal evidence.
Metadata Remover
Remove EXIF, IPTC, and C2PA metadata from any image in one pass.
AI Watermark Detector
A second opinion on the same file: SynthID plus a pixel classifier, with the limits of each.
Image Metadata Viewer
Read the provenance records a file carries before deciding which layer to clean.
AI Metadata Remover
Strip C2PA, IPTC and generator fields after a check confirms they are present.
Nano Banana Watermark Remover
Handles the full Nano Banana stack in one pass, and shows the result next to the original before you download.
All tools
Every remover and detector on one page, grouped by what each one cleans or checks.
Check a file for SynthID watermarks
The report separates proof from suggestion for each signal it reads. If your file carries declarations you want gone, the SynthID remover strips them and hands back a cleaned file.