Verification field guides
How to tell if a photo is a deepfake: six checks that hold up
To tell if a photo is a deepfake, check where it came from before how it looks. Find the earliest copy, read the file’s own records and labels, test the claim against independent sources, and treat visual clues as the weakest evidence, because good fakes hide them and real photos can show them.
Sources and dates
4
Reviewed
Put the next question to the file.
1. Find the earliest copy and who posted it
A deepfake usually travels as a repost or a screenshot. Search for the image with a reverse image search and note the oldest result you can find, the account that posted it and the date. Google’s About this image shows when Google first saw an image and where else it appears, which often surfaces the original or an earlier debunk.
If the earliest copy comes from a parody account, an AI art community or an account created days ago, that context matters more than any detail in the pixels.
2. Read the file’s records and AI labels
Ask for the original file, not a screenshot. Some generators and cameras attach records: C2PA Content Credentials can show who signed the file and which tools edited it, and many generators write an IPTC digital source type or saved prompt. A free checker such as the isGenAI image detector reads those records and names the field each finding came from.
Remember the limit: social platforms and screenshots often strip records, so a clean copy proves nothing either way. Google’s SynthID watermark can only be checked with Google’s own tools.
Sources: Coalition for Content Provenance and Authenticity: C2PA and Content Credentials Explainer, version 2.2Google DeepMind: SynthID
3. Test the claim against the world
A real event leaves other traces: other photographers, video from another angle, reports from outlets that were there. If a dramatic photo of a public figure exists in only one copy, the story is unverified whatever the pixels show.
Compare details with the claimed place and time: weather, signage, uniforms, the person’s known schedule. NIST’s overview of synthetic content notes that provenance, detection and context checks work best together, not alone.
4. Only then, look closely at the image
Zoom in on hands, ears, teeth, jewelry, text and reflections, and on the boundary between hair and background. Mismatched earrings, warped lettering or lighting that falls in two directions are reasons for suspicion. Newer models make fewer of these mistakes, and compression can create similar artifacts in real photos, so record them as clues, not proof.
- Save the copy you were sent and note where it came from.
- Run a reverse image search and record the earliest result.
- Check the original file for Content Credentials, AI labels and saved prompts.
- Look for independent coverage of the same moment.
- Note visual anomalies last, and state what remains uncertain.
Questions about this case
Can you tell if a photo is a deepfake just by looking at it?
Not reliably. Visual flaws such as odd hands or warped text are clues, but recent models avoid many of them and compression adds similar artifacts to real photos. Source, records and independent coverage are stronger evidence.
Is there a free tool to check if a photo is a deepfake?
Free checkers can read the evidence a file carries. The isGenAI image detector reads Content Credentials, AI labels and saved prompts from the original file and shows where each finding came from. No tool can prove a stripped copy is real.
What if the photo has no metadata at all?
Missing metadata is normal for screenshots and social media copies, so it proves nothing. Look for the original upload, other angles of the same moment and the account that first posted it.
Sources and dates
- Learn more about an image — Google Search Help.
- C2PA and Content Credentials Explainer, version 2.2 — Coalition for Content Provenance and Authenticity.
- SynthID — Google DeepMind.
- Reducing Risks Posed by Synthetic Content: An Overview of Technical Approaches to Digital Content Transparency — National Institute of Standards and Technology, 20 November 2024.
Sources checked on 20 October 2026. Linked reporting and official statements establish the attributed facts; the review steps are isGenAI’s analysis.
Put the next question to the file.
Choose the supported tool that fits your question. Keep the original source and save the output with your notes. Frame extraction, file-record inspection and still-image checks answer different questions; none alone authenticates a complete video or its audio.
Our methodology explains supported checks and result limits. Found a factual error or a newer primary source? Use our corrections process and include this article’s URL.
Keep checking: how to check a suspected deepfake, AI image and deepfake statistics and what X strips from uploaded images.