Start here
The free tool below runs entirely in this browser — no account, nothing stored or transmitted. Redact anything you would not want kept before you type it.
Open the companion tool →Why the eyes are the weakest tool
Single visual tells — extra fingers, bad shadows, unnatural blinks — are a moving target. Generators fix them faster than checklists update. What stays reliable is provenance: did this clip surface anywhere trustworthy, does the audio match the video, and does an independent source confirm the event.
The reshare decision is the actual stakes. A clip that cannot be traced to a source, corroborated, or matched to a real event should be held back — not because it is definitely fake, but because you cannot verify it and your followers cannot either.
The protocol
- Paste a description of the clip, its claimed source, and what it shows (no personal details).
- Read the provenance reading: origin, consistency, and corroboration each get a line with their basis.
- Before resharing, search for the event on an independent outlet; if none confirms it, hold the clip.
- If the clip shows a real person being impersonated, report it on the platform and flag it for the person affected.
What the tool does
Walk a video clip through a provenance-and-physics reading with the basis shown and a reshare decision. It runs in this browser; nothing is sent or stored.
Limit first: This check can raise your confidence, but a clean result doesn't prove a video is real. No single tell is reliable, and this tool does not analyze the video itself — it structures the provenance questions.
Open the companion tool →One human next step
Search the event on an independent outlet before sharing. Report suspected synthetic media to the platform, and see CISA's disinformation tactics guide for the pattern library.
Research log and safety checks
AI-assist path: a live acoustic/visual model pass is the canonical AI assist for media forensics, but is not wired in this static artifact; the walkthrough holds provenance checks the reader can run without a model, and the gap is named rather than papered over. §6.10: not a frontier-pace topic.
§0.5 protection result: category is deepfake-media-verification; crisis-adjacent: no; fear/urgency/scarcity toolkit used: no. The protective function is free and the paid feature is convenience only.