AI Image & Video Detector
Scan a photo or video for signs it was AI-generated — on your device.
Runs entirely in your browser — nothing you enter is uploaded, logged, or tracked.
Drop in a photo or video and this tool looks for signals that it was made by AI — Content Credentials and AI metadata first (the only near-certain clues), then image-artifact checks. It deliberately shows you the evidence and a confidence range instead of a yes/no verdict, because AI detection is genuinely unreliable: modern generators routinely beat every detector, and ordinary photos can trip false alarms. It runs entirely in your browser — your file is never uploaded.
Analyzing…
Frequently asked questions
Can this reliably tell me if an image is AI-generated?
No, and any tool that claims it can is overselling. On-device detection is weak: today’s generators (Flux, Midjourney v6, DALL·E 3, Sora) routinely evade detectors, and real photos can be falsely flagged. The one near-certain signal is provenance metadata — Content Credentials (C2PA), an IPTC “trained algorithmic media” tag, or Stable Diffusion parameter chunks. If those are present we say so with high confidence; otherwise we show weak signals and a confidence range, never a definitive verdict.
How does it work without uploading my file?
Everything runs in your browser. Metadata and Content-Credentials parsing is plain JavaScript on the file’s bytes; the artifact checks use the canvas. Your photo or video is never sent anywhere — you can even run it offline after the page loads.
What does “no signs found” mean?
It means we found nothing that looks AI-generated — which is NOT proof the file is real. Metadata is easily stripped, and good AI images often leave no detectable trace. Treat a clean result as “we couldn’t tell,” not “confirmed authentic.”
Why won’t it just give me a single percentage?
A single confident-looking percentage would be misleading. We show the evidence layer by layer — provenance, then artifact checks — with their weights, plus a rough range, so you can judge for yourself. A point estimate alone hides how uncertain the answer really is.
Should I use this to prove someone used AI?
No. Please don’t use it to accuse anyone, or for hiring, academic, legal, or forensic decisions. False positives are common and harm real people. It’s a curiosity and triage aid — verify the source and consult a human expert for anything that matters.
How does the video check work?
It samples several evenly-spaced frames, checks each for AI signals, and reports the spread. It cannot analyze audio, lip-sync, or motion, and it analyzes each frame on its own — so it’s even less reliable than the image check. Temporally-smooth deepfakes will usually pass.