Gemini← All featuresSynthID & verification
SynthID · AI-media verification

Check whether media used Google AI

Gemini can inspect an uploaded image, video or audio file for Google SynthID and compatible Content Credentials. These signals help identify origin or editing history, but they do not prove whether a claim is true.

Verification currently has file-size, duration and usage limits.
Two signals answer different questions
〰️SynthIDChecks for Google’s invisible watermark in media made or edited by supported Google AI models.
📜Content CredentialsShows compatible provenance and edit-history information supplied by apps or devices.
🧠Your judgmentCombines the signal with the source, content, metadata and other evidence.

How to read a SynthID result

Result
What it means
What it does not mean
Detected
All or part of the media was created or edited by supported Google AI.
It does not confirm the truth of the content.
Not detected
Gemini did not find a Google SynthID signal.
It does not prove the media is human-made or non-AI.
Unclear
The file may have too little watermarkable detail or a weakened signal.
It is not a reliable conclusion about origin.

Check one file deliberately

📎

Use the original file when possible

Upload one supported image, video or audio file rather than a collage or heavily compressed screenshot.

Ask a precise question

Ask whether the file was created or edited by Google AI instead of whether it merely “looks AI-generated.”

🧭

Interpret both signals

Read the SynthID result and any compatible Content Credentials information before concluding.

Non-detection is not proof of human origin

✂️

Edits can weaken a watermark

Resizing, recolouring and compression may leave it detectable, but enough alteration can remove the signal.

🌐

Other AI systems are outside SynthID

Gemini’s check is not a universal detector for every company’s generative media.

When the decision matters, use more evidence

🔎

Inspect and reverse-search

Look for visual inconsistencies and search for earlier versions or the original source.

🧾

Review metadata and credentials

Use available provenance information as supporting evidence, not as a universal truth certificate.

⚖️

Make a combined judgment

Consider the signal, source, context and consequences together.

Know the practical limits

📦

One file at a time

Current guidance allows one file up to 100 MB, with additional video and audio duration limits.

🎞️

Detection can cover part of a video

A highlighted portion does not necessarily mean every frame contains the watermark.

Provenance is not factual accuracy

Neither SynthID nor Content Credentials tells you whether the media’s underlying claim is true.

💡
Practical rule

Detection is meaningful evidence of Google AI involvement. Non-detection is only the absence of a detected Google SynthID signal.