Google's provenance tools
Google says SynthID embeds an invisible watermark in media generated by its AI tools and that it has watermarked more than 100 billion images and videos. In the Gemini app you can upload an image, video or audio file and ask “Is this made with AI?” to check for that signal. Google has also added C2PA Content Credentials verification to Gemini, which can show whether a file records a camera capture or the tools used to edit it. These are Google's services; Image Evidence has no SynthID connection.
Source: Google: identifying AI-generated media (opens in a new tab)
Generation, editing and ordinary adjustments
The question you ask shapes the answer. A newly generated scene, a camera photo with an inserted object and a photo with ordinary exposure adjustments have different histories. Ask for the original capture and a description of the edits when the owner can provide them. An AI-related signal does not mean every pixel is artificial, and an overall score cannot locate an edit or estimate its size.
Choose the strongest input
Prefer the original export to a screenshot or thumbnail, because screenshots drop Content Credentials and can weaken other signals. Keep the source link and the creator's disclosure in your own notes so any result keeps its context. Video adds complications: platforms transcode uploads, and a short clip may mix generated and recorded material.
What a result can and cannot show
A SynthID detection supports the conclusion that Google AI was used in making the file; it does not show that the depicted event happened or that a caption is true. A C2PA manifest is meaningful only after its signature and signer are validated, and an ordinary software field can be edited by anyone. Image Evidence has not completed a Gemini benchmark; a fair evaluation would report false positives on camera images as well as detection of generated and edited examples.
Review beyond the detector
For a watermark question, use Google's official tools and review their upload terms. For a factual claim, compare the image with independent evidence of the event, because even a verified origin signal would not establish that the scene is accurate. Treat results as information, not a forensic opinion, and avoid consequential decisions about students, applicants, creators or other people based on a detector label alone.
How do I check if an image was made with Gemini?
Upload the original file to the Gemini app and ask “Is this made with AI?”. Gemini checks for Google's SynthID watermark and can also read C2PA Content Credentials. If you only have a screenshot, ask for the original, because a screenshot drops Content Credentials and can weaken other evidence.
Can Gemini detect images made by other AI tools?
Gemini's check looks for SynthID watermarks and for Content Credentials it can read. Google says other companies, including OpenAI and ElevenLabs, are adopting SynthID, but a “not found” result still does not prove an image is real: many generators use neither signal, and edits can remove metadata.
Does a SynthID result prove where an image came from?
No. It is evidence that a SynthID-watermarking tool was involved, not a complete history. It does not tell you who made the image, what was changed, or whether the image is being presented honestly. Combine it with the source, the claim and any other provenance records.