Preserve the full clip
Keep the earliest copy with its audio and surrounding footage. Cropping, screen recording and re-encoding can remove details needed to interpret the result.
Loading…
Upload a clip up to 200 MB in MP4, MOV, WEBM, MKV, AVI or another supported video format. The report separates generated visuals, face manipulation and synthetic voice signals so one weak signal does not become an overconfident verdict.
Keep the earliest copy with its audio and surrounding footage. Cropping, screen recording and re-encoding can remove details needed to interpret the result.
Generated visuals, face manipulation and synthetic speech answer different questions. A clip without a usable face or voice cannot provide all the same evidence.
Compare the original publication and trusted recordings. An uploaded clip check does not verify a live call, identify the speaker or confirm their request.
Select short MP4, MOV or M4V clips in your company workspace: up to 20 MB and 60 seconds per video. Review each result and download individual reports with a CSV summary.
Batch video checksConnect the API to check video uploads in your application. Submit one job per video and track completion through job status or signed webhooks. API uploads have their own file limits.
AI image & video detection APILooks for frame-level patterns associated with synthetic video.
Reviews visible faces for manipulation and face-swap indicators.
Checks the soundtrack when usable speech is present.
The highest available model score informs the headline, while the individual checks remain visible. A low score means no strong signal was detected in this clip; it does not certify that every frame is authentic.
Important limitation: Very short clips, heavy compression, screen recordings, animation and videos without a clear face or voice can reduce reliability.
Read the full methodologyDemonstration only: this is not a customer case, live result or performance claim.
A reposted clip appears to show a real person making an urgent claim, but the original upload is unavailable.
The report keeps generated-video, face-manipulation and synthetic-speech signals separate, including when they disagree.
A suspicious signal supports further review. It does not identify a creator or establish that the surrounding claim is false.
Locate the earliest full-length copy, compare key frames and speech with trusted sources, and verify the apparent speaker independently.
Check a surprising clip before reposting it.
Review unfamiliar footage when identity matters.
Add a detector result to source and provenance checks.
Use the earliest, highest-quality clip available. Check its source and context, then use the AI video detector to review generated-visual, face-manipulation and synthetic-voice signals.
No. A low score only means the tested models did not find a strong signal in that copy. Provenance, source history and the claim around the clip still need verification.
AI video includes fully generated scenes, objects and motion. A deepfake more narrowly imitates or alters a real person, often through face replacement, lip-sync or synthetic voice. Use the dedicated deepfake video page when impersonation is the main concern.
Use source history, provenance and independent checks before making a high-stakes decision.
Read the AI-video verification guide