One convincing frame doesn’t explain a whole clip. Inspect source evidence, follow visual signals across sampled moments, and see what was checked in the soundtrack.
Up to 2 minutes Private upload
Evidence first, even when the answer is uncertain. Passive visual findings remain Inconclusive during beta. Basic Scan does not classify AI voices or face swaps.
Upload the original if you have it. We identify the media and show the matching report.
Drop an image or video hereor from your deviceJPG, PNG, WebP, AVIF, HEIC/HEIF, BMP, TIFF, GIF, APNG · Up to 20 MBMP4, MOV, WebM, M4V, MKV, AVI, MPEG, 3GP · Up to 200 MB / 2 min
No files selected.
Loading secure verification…
Uploaded originals are kept privately for future analysis until you delete them. The report expires separately. Use “Delete uploads and report” on your scan page. Privacy details.
Private upload. Keep this browser’s access token to manage deletion.
MP4 · MOV · WebM · M4V · MKV · AVI · MPEG · 3GP Up to 200 MB / 2 min. Images work here too. Have a Live Photo? Check still and motion.
FROM FILE TO REVIEW WINDOW
A clip has a timeline. So does the evidence.
A generated insert can sit between ordinary footage. Sampling across the clip and revisiting selected moments helps put individual findings in context.
ANATOMY OF A 60-SECOND CLIPIllustration · not a scan result
00s
05s
10s
15s
20s
25s
30s
35s
40s
45s
50s
55s
00:0000:3001:00
Spaced samples across the clip
Nearby frames revisited
Illustrated review window: 00:30–00:40. Highlighting means “look here,” not “AI confirmed.” Unsampled moments remain outside the visual check.
01
Read the whole file first
The scan checks the container, duration, codec and tracks, then inspects available Content Credentials on the video itself. Export settings give context; only a valid, trusted AI declaration can support a strong source finding.
02
Inspect moments across the clip
Visual checks inspect sampled moments in the video. Findings help direct attention to parts of the clip that need review, while the report explains the limits of that coverage.
03
Return to the moments that matter
Time segments help you locate sampled findings in your original clip. Review the surrounding action and its source before drawing a conclusion. A segment is a review window, not an exact boundary around an AI edit.
Sampling is a trade-off. It keeps a short-video scan practical, but fast cuts, subtle edits and moments between samples may be missed. Available timestamps describe the uploaded clip; they do not certify the full recording.
TWO TRACKS. SEPARATE QUESTIONS.
The picture can be real. The voice can be synthetic.
A video combines visual content and a soundtrack. Neither can stand in as proof for the other, so their coverage is reported separately.
VISUAL TRACK
What was shown?
Source declarations and sampled frame signals help investigate generation. They do not verify a person’s identity or automatically identify a replaced face.
Source credentials & frame signals
Basic checks
Specialized face-swap classification
Not in Basic
Calibrated visual probability
Pending
AUDIO TRACK
What was heard?
Track information and available activity checks describe the audio present. Speech activity, silence or a cut is not proof of a cloned or generated voice.
Track information & activity
Basic checks
Synthetic / cloned speech classification
Not tested
Video without an audio track
Not applicable
Deep Scan is not publicly available yet. An unavailable or failed check is a coverage gap, not a negative finding.
KEEP THE CONTEXT IN FRAME
Before you upload that clip.
Choose the file that best preserves the moment you want to investigate. Re-exporting can change both the evidence and the meaning of its timestamps.
Start with the original download
Screen recordings, messaging apps and repeated exports may strip credentials and soften frame patterns. Keep the source URL, caption and publication date alongside your file for your own review.
Keep the lead-in and what follows
If you need a shorter excerpt, include the action before and after the moment in question. Record where it begins in the original. Cutting tightly around one frame can hide relevant context.
Leave the soundtrack attached
Removing or replacing audio changes the material under review. Preserve the version you received, even if the voice sounds suspicious. Basic audio inspection cannot settle whether a speaker is synthetic.
SHORT-VIDEO CHECKING, EXPLAINED
AI video detector FAQs
What frame sampling can tell you, where deepfake detection needs more coverage, and how to handle clips shared online.
How do I check if a video is AI-generated?
Upload a supported video file and run Basic Scan. Read the source evidence, frame coverage and time segments together. During beta, uncalibrated visual signals remain Inconclusive. A trusted credential declaring AI generation may support a stronger finding. Also trace the original publisher and compare the clip with its claimed context.
Does the AI video detector inspect every frame?
No. Video analysis uses sampled moments rather than checking every frame. Very short edits and unsampled moments can be missed, and completing a scan does not mean the entire recording has been verified. Read the findings together with the reported coverage and limitations.
Can it detect a face swap or a deepfake video?
Basic Scan does not run a specialized face-swap detector or verify a speaker’s identity. It may find a trusted declaration of AI editing, but sampled whole-frame signals cannot rule out a replaced face. Specialized manipulation checks are reserved for Deep Scan, which is not publicly available yet.
Can a real video contain an AI-generated voice?
Yes. The picture and soundtrack can have different origins. Basic Scan checks available audio-track information and activity, but does not classify synthetic or cloned speech. AI voice coverage is reported as not tested when that classifier has not run. A visual finding cannot authenticate the soundtrack.
Can I check clips made with Sora, Veo or Runway?
You can submit a supported video regardless of its claimed generator. This beta does not have published, generator-specific accuracy results and cannot promise to identify a particular tool or model version. Vendor watermarks that require an unavailable official verifier are outside the scan’s coverage.
Can I paste a TikTok, YouTube or Instagram link?
Not currently. Upload a video file from your device that you have permission to inspect. Social-platform downloads, subtitles and re-encoding can change visual signals or remove source credentials. Keep the publication URL for your own source research, and use the original file when it is available.
What are the video upload limits?
MP4, MOV, WebM, M4V, MKV, AVI, MPEG and 3GP files up to 200 MB and two minutes are accepted. Duration, decodability, frame dimensions and other processing limits are checked after upload; a supported extension does not guarantee that its codec or contents can be processed. Longer videos and live streams are not supported. If you export a shorter clip, retain the original and record the excerpt’s start time.
What does a flagged time segment actually mean?
It points to a sampled moment or group of moments worth reviewing. It does not establish an exact edit boundary, identify the editor or prove that everything outside the segment is genuine. Timestamps refer to the uploaded file, so an excerpt’s times will differ from those in the full recording. Uncalibrated frame scores are not measured AI probabilities.
What happens to my video after the scan?
Original uploads are retained privately for future analysis until you delete them. Reports expire after 45 minutes, independently of the originals. Keep the scan URL and this browser’s access token to use Delete uploads and report, including after expiry. Basic Scan does not require an account, and usage limits may apply.
BRING THE WHOLE MOMENT
What’s behind that clip?
Start with one video. Follow the evidence through time.