Know what is real before you publish
Metadat-AI checks every video for signs it was made or changed by AI, inside your own environment, so you can label it, check it or stop it before it reaches your audience.
Illustrative example. Each frame is scored from 0 (real) to 1 (synthetic).
Audiences want to know what is real
AI video now looks real. Viewers, advertisers and regulators expect you to say when it is not.
- It is now the law in the EU. Since 2 August 2026, the EU AI Act requires AI-generated or manipulated video that could pass as real to be clearly labelled when it is published.
- The fines are real. Failing to label can cost up to €15 million or 3% of global annual turnover, whichever is higher.
- Trust is hard to win back. One unlabelled deepfake can undo years of audience trust.
You cannot label what you have not found. That starts with checking every video.
- 2 August 2026EU AI Act labelling rules for deepfakes apply from this dateSource: EU AI Act, Article 50
- €15 millionor 3% of global annual turnover, whichever is higher, is the maximum fineSource: EU AI Act, Article 99
Spots what the eye misses
The detector looks for the low-level traces that AI video generators leave behind, not for what is in the picture. That means it works across different AI video generators and stays robust to everyday compression and re-encoding. It runs alongside your Metadat-AI tags, in one pipeline.
- Up to 92%accuracy on uncompressed videoSource: Detection model published results
- 87%accuracy even at 15% compression
- 82%accuracy even at 50% compression
How it works
- Every frame is checked Each frame is scored from 0 (real) to 1 (synthetic), and the whole video gets one overall score
- Flagged with a timecode High scores are flagged at the exact moment, so reviewers go straight to the section that matters
- People decide Your team reviews the flag, then labels, checks or removes the content
Where it fits
- Upload screening Score new uploads before they go live
- Archive audit Check existing libraries for synthetic content at scale
- Licensing checks Confirm footage is authentic before you license or publish it
- Newsroom verification An editorial check on user-generated and social footage
- Impersonation Flag synthetic video of executives, officials and talent
- EU AI Act labelling Catch AI content in time to label it before it is published
Runs in your environment
The detector is deployed into your own cloud account or onto your own hardware, for example a GPU instance in your AWS account. Video is analysed where it already lives, and scores are written to your own metadata and logs. Your video never leaves your environment, so your existing security controls stay in place.
Human in the loop
The detector gives a likelihood, not a verdict. Trained reviewers make the final call. Metadat-AI does not make identity, age or consent decisions, and this is not legal advice.
FAQ
What does the EU AI Act require?
Article 50 requires anyone who publishes AI-generated or manipulated video that could pass as real (a deepfake) to disclose it clearly, at the latest when people first see it. Evidently artistic, satirical or fictional work only needs a light-touch disclosure. The rules have applied since 2 August 2026.
Is there a set wording for the label?
No. The law requires a clear disclosure. The EU has published optional icons for fully AI-generated and partially AI-modified content, and recommends adding a short, plain-language text label.
How accurate is it?
The detection model's published results show up to 92% accuracy on uncompressed video, 87% at 15% compression and 82% at 50% compression. Heavy compression and re-encoding can lower it, so we set the flag threshold with you on your own content.
Which formats does it support?
MP4, MOV, MKV and WebM video, with H.264, H.265, AV1, VP8 and VP9 codecs. Audio is not analysed.
Does it say which AI tool made the video?
No. It scores how likely the video is to be AI generated or changed, frame by frame.
What hardware does it need?
It runs on GPUs, in your own cloud or on your own hardware. We size and set it up for you.
Can it be added to Auto-Compliance?
Yes. AI video scores sit alongside your compliance flags and tags in one pipeline.