Deepfake Detector

The deepfake detector is tuned against Veo 3.1, Kling 3.0 and Midjourney v8. New models are added to the ensemble within 30 days of release.

We benchmark at 99.1% accuracy on the RAID benchmark with a 1.2% false-positive rate for text, and 96.9-98.4% across image, audio and video. Every verdict ships with a confidence band so you know how certain the result is.

Edited prose, technical writing and ESL writing are common false-positive triggers for naive detectors. We tune against those failure modes specifically and surface confidence so a low-certainty AI verdict reads differently from a high-certainty one.

Yes. Every plan keeps every scan forever - Free included. There is no retention tier and never will be. You can delete an individual scan from /dashboard/history at any time.

No. Uploads are processed in-memory and never used to train detectors. We are SOC 2 in progress and ship a Data Processing Addendum.

Every signup gets $1 in scan credit, usable across every detector model - text, image, audio, video. There is no expiry while your account is active.

Yes - Starter handles 10 files at once, Pro 50 and Business 500. Drop a folder into the dashboard and we'll scan everything in parallel.

Yes - included on the Business plan. The API mirrors the dashboard, supports streaming results for long videos and ships an OpenAPI spec.

We don't sell humanisers, evasion or AI-paraphrasing. We're a verdict tool. If you want to bypass detection, this isn't the product for you.