Disassemble the video. Return JSON {"visual_signs":[...],"context_checks":[...],"verdict":"..."}: visual_signs - signs of fake in the video itself; context_checks — checking the context and source; verdict - an informed decision on what to do. Return only JSON, no text around it.
Mechanics: Generative models create faces, voices and videos that never existed. There is no ideal - there are traces. Visual signs: strange fingers, teeth, ears, asymmetrical earrings and glasses, “floating” hair, distorted text in the background, unnatural blinking, lips out of sync with sound, too smooth, waxy skin. But the signs are quickly becoming outdated - today the main method is not to “look at the pixels”, but to check the CONTEXT. Technique: reverse image search and the question “where is the original source?” - a fake usually pops up without a story, only in transfers, and it is not on the person’s official channels. The second technique against voice deepfake: call back a known number and ask a question, the answer to which only a real person knows - the “family code word.” Insider: it’s not the picture that gives away the scammer, but the scenario—urgency, secrecy, and a request for money or codes. A real friend and a real bank will wait for verification. A typical mistake: believing a video or audio just because “his face and voice are there” - this is what is now most easily faked.
Unlock access to submit solutions for instant AI review.