To check an image for a Google SynthID watermark, use Gemini's official verification workflow. AIFakeScan does not detect SynthID. Its image scan can help you inspect other available evidence, but a visual model result is not a watermark result.
This guide separates the provider check from the broader question of whether an image is trustworthy. Official product instructions were reviewed on September 25, 2026. The worksheet below is a suggested review method, not a benchmark or a report of detection accuracy.
Where can you check SynthID?
Google's SynthID overview directs users to Gemini for verification. It also describes a separate SynthID Detector portal being tested with journalists and media professionals, with an early-access waitlist. Do not assume that portal is available to every account.
SynthID embeds an imperceptible watermark in generated media. Google describes it as designed to withstand some transformations, including cropping and lossy compression. That design goal does not establish a success rate for the particular file in your hands.
How to verify an image in Gemini
Follow Google's current Gemini verification instructions. Sign in with an eligible personal or Workspace account, open Gemini, use Add files, and upload one image. Ask whether Google AI created or edited it. Read the returned verification explanation and preserve the exact response alongside the file.
Google currently specifies a 100 MB maximum for a single verification upload. Check the help page for changes and account requirements. These are Gemini limits, not AIFakeScan upload limits.
If only a screenshot is available, Google advises a tight crop around one image rather than a collage. Keep the untouched screenshot as well, and record the crop. Request the original download when possible.
What does the result support?
Gemini's help distinguishes a detected SynthID watermark, no detected watermark, and cases where the determination is unclear. It currently limits Gemini's SynthID recognition to Google AI content, even though other companies have adopted SynthID. A positive result can concern all or part of an image, including editing.
Use this interpretation checklist; the labels below are review categories, not promised interface wording.
| What you receive | What to record | What to do next |
|---|---|---|
| A reported SynthID detection | Exact response, file, date, and any described scope | Distinguish generation from editing; verify the image's caption separately |
| No reported detection | The negative finding for this submitted file | Keep other generators and the image's source history in scope |
| An unclear determination | The explanation without converting it to a yes or no | Seek a better source file and retain the uncertainty |
| Upload failure or no completed check | Error or missing coverage | Mark it as not checked; do not record a negative result |
A detector response should not silently become a broader claim about a person, event, or publisher. For example, finding evidence of an AI edit does not identify every changed region or establish that an accompanying news claim is false.
SynthID, Content Credentials, and visual detection
These evidence channels can appear in the same review. Keep their outputs separate.
| Evidence channel | Question to ask | Limit to preserve |
|---|---|---|
| SynthID verification | Did a compatible provider check find its watermark? | A watermark check has defined provider coverage |
| C2PA Content Credentials | What provenance declarations were verified, and under which trust rules? | Signed history is not a guarantee of factual truth |
| Ordinary metadata | What camera, software, or export clues remain? | Editable fields are context, not independent proof |
| Visual model | What did this model report about these submitted pixels? | A classification does not reconstruct creation history |
Gemini can also inspect Content Credentials. Save those findings separately from its SynthID response. Our C2PA guide explains validation and trust states.
A worksheet for originals and shared copies
Keep one record per file. If you compare an original with a social download or screenshot, label the versions before uploading. Do not overwrite an original while trying to obtain a different detector answer.
| Field | Fill in during your review |
|---|---|
| Review question | The specific claim about generation, editing, or source |
| File identity | Filename, format, dimensions, and optionally a SHA-256 hash |
| Acquisition | Source URL or sender, download date, and whether it is an original or copy |
| Known changes | Crop, screenshot, resize, export, or platform route; unknown if unavailable |
| Official verification | Service, date, exact response, and whether the check completed |
| Other evidence | Credential details, publication history, creator explanation, and conflicting findings |
| Conclusion | What the evidence supports and what is still unknown |
This is an empty review template. It contains no measured watermark survival results. A comparison of two unknown files also cannot tell you which signal existed before either file reached you.
Where AIFakeScan fits
Use the AI image detector to inspect available file evidence and read the coverage notes. AIFakeScan does not run Google's proprietary SynthID verifier. Its beta visual signals remain Inconclusive on their own and do not provide a calibrated AI probability.
For a broader source review, follow How to Detect AI-Generated Images. If a detector flag conflicts with the source history, use the false-positive review guide.
Use the interactive SynthID verification route
The AIFakeScan SynthID checker turns the workflow in this guide into a short decision route. Choose the suspected source, whether you have an original or screenshot, and whether you need to check creation or editing. It then provides the current official Gemini route and a prompt you can copy.
The route does not upload a file to AIFakeScan and does not decode SynthID. The result comes from the official verifier you open, so record the response there as detected, not detected or unclear.
This guide remains the detailed reference for coverage, file-version handling and result interpretation. The tool page is the operational entry point for `synthid checker` intent.
Frequently asked questions
Is a general AI image detector a SynthID checker?
No. A visual classifier and a compatible watermark verifier inspect different evidence. Look for an explicit description of the check that actually ran.
Does no SynthID detection prove an image is a real photograph?
No. The result does not rule out every other AI system or establish the truth of the depicted scene. Record the narrow finding and continue the source review.
Can AIFakeScan verify SynthID?
No. Use Google's official Gemini workflow for that check. AIFakeScan provides complementary evidence inspection with its own stated coverage limits.
Can a screenshot be reviewed?
Yes, but preserve it as a separate file version and seek the original. Do not treat a result about the screenshot as a complete history of the unseen original.
Sources
- Google DeepMind: SynthID — Google DeepMind
- Verify AI-generated images, videos, and audio — Google
- C2PA and Content Credentials Explainer — C2PA