AI Video Detector
Upload a video to screen container metadata, keyframe spatial texture, and C2PA / Content Credentials markers locally in your browser.
Clips stay in your browser. Nothing is uploaded to a server.
How AI Video Detection Works
EyeSift reads the selected video locally and screens file-level signals such as dimensions, duration, estimated bitrate, aspect ratio, resolution, model-friendly sizing patterns, and C2PA / Content Credentials marker presence. These signals can help prioritize manual review, especially for short social clips, but they are not a forensic frame-by-frame verdict.
C2PA and Video Provenance
C2PA Content Credentials can travel with some video files and editing/export workflows. EyeSift checks for C2PA-style markers locally as a fast provenance clue. A marker is not the same as full signature verification, and missing markers do not prove a video is AI-generated because platforms often strip metadata.
Limitations
Video AI detection is an emerging field. Results should be considered preliminary. Very short clips, screen recordings, exports from editing apps, or heavily compressed videos may produce less reliable results.
Provenance Guides
Deepfake Detection Guides
Related Tools
Which AI video generators does this cover?
EyeSift is generator-agnostic. It screens file-level signals such as resolution, duration, estimated bitrate, aspect ratio, model-friendly dimensions, and C2PA marker presence, so it can be used as a first pass on clips from Sora, Runway, Kling, Google Veo, Pika, Luma, and face-swap deepfake apps. It does not analyze individual frames, and file-level heuristics alone cannot say which generator produced a clip.
Attribution becomes possible when provenance survives: OpenAI attaches C2PA Content Credentials to Sora exports and Google links Veo output to SynthID, and EyeSift flags C2PA markers in the file when they are present so you can open them in a dedicated Content Credentials verifier. Re-encoding by social platforms and messaging apps usually strips them, so a missing credential is never proof that a video is authentic.
How this compares with other AI video detectors
| Detector type | Where your video goes | What you get | Best for |
|---|---|---|---|
| EyeSift (this page) | Stays in your browser; no upload, no account, no size-based pricing | AI-risk score, C2PA / Content Credentials marker check, and file-level metadata checks (resolution, duration, bitrate, aspect ratio) | Private first pass and provenance check before you share or escalate |
| Server-side deepfake classifiers | Uploaded to the service; length caps, credits, or logins are common | Percentage verdict from a hosted model, sometimes per-frame or face heatmaps | A second opinion when privacy is not a concern |
| Enterprise media forensics (Reality Defender, Hive, Sensity) | Processed on their infrastructure under contract | Multi-model detection and API access; accuracy figures are vendor-reported under their own test conditions | Newsrooms, platforms, and fraud teams that need defensible results |
Published deepfake-detection research has found that detectors often lose accuracy on re-encoded, cropped, or low-resolution clips, so treat any single score, including this one, as a screening signal rather than proof.
Frequently Asked Questions
Can I record a clip with my phone camera instead of uploading?
Yes. On mobile, use the camera button under the upload box. The clip is analyzed on your device like any upload and is never sent to a server.
Can it tell which AI generator made a video?
No, not by itself. EyeSift flags C2PA / Content Credentials markers when they are present, and a dedicated verifier can show which tool signed them, but EyeSift does not read generator names from video files. File-level heuristics alone cannot attribute a clip to Sora, Runway, Kling, or Veo.
Does it work on clips downloaded from TikTok, Instagram, or WhatsApp?
It can analyze them, but platforms re-encode video and strip provenance data, so the check relies on file-level heuristics alone and should be treated as weak evidence. Request the original export whenever possible.
What AI video generators can this tool detect?
EyeSift can screen videos that may have been produced by tools such as Sora, Runway, Pika, Kling, and similar generators by checking local file, C2PA marker, and metadata clues. It does not claim forensic certainty; use the score to decide whether the video needs provenance checks and manual inspection.
Does EyeSift verify C2PA Content Credentials in video files?
EyeSift checks for C2PA / Content Credentials markers in the local video file as a provenance clue. It does not perform full cryptographic signature validation in the browser, so use a dedicated Content Credentials verifier when the result matters.
How does deepfake video detection work?
Practical deepfake review combines metadata, provenance, source credibility, visual inspection, and specialist forensic tools. EyeSift provides a browser-side triage score from available file signals such as resolution, duration, bitrate, aspect ratio, and generation-friendly dimensions.
What video formats and lengths are supported?
We support MP4, WebM, and MOV formats. Very short, heavily compressed, screen-recorded, or re-exported videos may produce less reliable results.
Can AI-generated videos pass detection?
Yes. AI video generation and editing workflows are improving quickly, and metadata can be stripped or changed. Treat EyeSift as an initial risk screen and combine it with provenance checks, source review, reverse search, and human visual inspection for critical verification.
Is video AI detection reliable enough for journalism?
AI video detection should be one verification tool among several in journalistic workflows. No automated detector is 100% accurate. Journalists should also check provenance, metadata, source credibility, reverse search, and expert review for high-stakes footage.
Is my uploaded video stored anywhere?
No. Video processing happens entirely in your browser using client-side JavaScript. Your video is never uploaded to any server. The video data exists only in your browser memory during the session and is released when you close the page or clear the upload.