What the AI video detector checks
| C2PA Content Credentials | Signed manifests in MP4/MOV (for example Google Veo) that declare digitalSourceType = trainedAlgorithmicMedia. The strongest evidence when the signature validates. |
|---|---|
| Container metadata | ftyp brand, udta/meta/ilst, encoder, ©too and software fields, AIGC labels and platform strings written by generators or their export pipelines. |
| SEI markers | Tool strings inside the H.264/H.265 bitstream that can survive a remux but usually not a re-encode. |
| Frame watermarks | When the container has nothing to say, frames from several timestamps are checked for visible corner watermarks and invisible watermark patterns. Decoded locally with ffmpeg.wasm. |
How to tell if a video is AI generated
- Use the original file: Download the MP4 or MOV straight from the generator or the person who made it. Screen recordings and social-media re-downloads usually lose the evidence.
- Drop it into the detector above: Parsing runs in your browser. The video is not uploaded to a server.
- Read the evidence layer by layer: Check the C2PA state first, then container fields with the box path they came from, then SEI and frame results.
- Interpret an empty result carefully: No signal means this copy carries no readable evidence. It does not prove the video was filmed by a person.
Want the full method with real samples? Read the guide to telling if a video is AI generated, or inspect the raw container with the MP4/MOV metadata inspector.
Video generators and the evidence they leave
| OpenAI Sora | C2PA and MP4 metadata on original downloads; a visible corner watermark on many exports, which the frame check can flag. |
|---|---|
| Kling AI | Container tool fields and frame marks on original MP4 files. |
| Runway | Encoder and container metadata; frame checks when fields are stripped. |
| Google Veo | Signed Google Generative AI C2PA manifest, verified on a real Veo MP4 sample. |
Why this is not a “deepfake score”
Pixel classifiers return a probability that is hard to explain and easy to fool with re-encoding. AICheck365 reports only evidence stored in the file — a signed manifest, a metadata field, a known watermark — so you can see why a result was reached and verify it yourself. The trade-off is honesty about gaps: when a platform strips the evidence, the detector says so instead of guessing.
FAQ
- Is this AI video detector free?
Yes. It is free, needs no account, and has no upload limit because the file is analyzed on your own device.
- How do I tell if a video is AI generated?
Start from the original MP4 or MOV. Check for a valid C2PA manifest that declares AI generation, then container fields and SEI markers left by the generator, then known frame watermarks. Visual inspection alone is no longer reliable for current video models.
- Can it detect Sora, Veo, Kling and Runway videos?
It recognizes the provenance signals these tools currently write — C2PA manifests, container metadata and known watermarks — when they are still present in the file. It is not a general classifier that guesses from pixels.
- Does it work on TikTok, YouTube or Instagram downloads?
Sometimes a watermark or file-name clue survives, but these platforms re-encode uploads and usually strip C2PA and container metadata. For a reliable answer, find the first published or original file.
- Is my video uploaded?
No. The container is parsed with WebAssembly in your browser, and frame extraction, when needed, also runs locally.