Is That Video Real or AI-Generated?

Upload the video to the Imagera AI Video Detector — MP4, WebM, or MOV up to 5 minutes. The detector runs six independent forensic methods (temporal consistency, frame-level frequency analysis, lip-sync cross-check, optical-flow physics validation, diffusion fingerprinting, and multi-signal fusion) and returns a verdict in seconds: AI probability from 0–100%, a confidence score, the suspected source generator, and a per-method forensic breakdown.

Deepfake and face-swap detection is measured at 96.1%. Full-video detection carries no published accuracy figure, because we do not have a measured one to publish — the report gives you a calibrated confidence score instead, and that is the number to weigh. Accuracy on heavily compressed, cropped or re-encoded footage is lower in every case. Treat any result as a likelihood, never a verdict.

Yes. The detector covers both cases: fully synthetic video from text-to-video and image-to-video generators, and partially manipulated real footage — face swaps, face re-enactment, and lip-sync dubbing from tools such as FaceSwap, DeepFaceLab, SimSwap, FaceFusion, Wav2Lip, and HeyGen. Deepfake-specific scans cost 15 credits; full video scans cost 20 credits.

Detection models are trained on outputs from 20+ video and face tools: Sora 2, Google Veo 3, Runway Gen-4 Turbo, Kling 3.0, Pika 2.5, Seedance 2.0, Luma Ray 2, MiniMax Hailuo, and Vidu Q3 for generated video, plus FaceSwap, DeepFaceLab, SimSwap, FaceFusion, Wav2Lip, LivePortrait, Reface, HeyGen, and SadTalker for deepfakes. New generators are added within days of public release.

MP4, WebM, and MOV files up to 5 minutes long. For longer material, scan the most critical segment — talking-head sections are the most information-dense for deepfake analysis. Typical scans complete in 10–30 seconds depending on length and resolution.

Yes, with a caveat. Temporal signals — blinking cadence, motion-field anomalies, lip-sync drift — survive re-encoding because they live in the content itself, not the file. Heavy compression does degrade frame-level frequency forensics, which is why the verdict fuses multiple independent signals and reports a confidence score rather than a bare yes/no.

Every scan returns a binary verdict (AI-generated or authentic), an AI probability percentage, a calibrated confidence score, a human-readable label such as "Likely AI-Generated," the suspected source model (for example a specific video generator or face-swap tool), and a forensic breakdown showing which detection methods triggered and why.

Pay per scan with no subscription. Full video scans cost 20 credits; deepfake and face-swap scans cost 15 credits. Credits are pay-as-you-go, and current pack sizes and prices are on the pricing page. No monthly fees or minimum commitments.

From 2 August 2026, Article 50 of the EU AI Act requires deepfakes and AI-generated media to be clearly disclosed, with penalties for transparency violations of up to EUR 15 million or 3% of global annual turnover. Detection is how platforms, publishers, and compliance teams operationalize that duty: scanning inbound video to flag undisclosed synthetic content before it is published or relied upon. Imagera reports provide the documented, per-method evidence trail those workflows need.

Your video is uploaded and stored in your account while it is analysed and afterwards, which is also what lets the scan appear in your history. It is not used to train models and not shared with third parties, and all transfers are encrypted with TLS 1.3. If you would rather it not be retained, delete the scan from your history when you are done.

Detection reports include confidence scores, methodology breakdowns, and suspected source-model identification that can support proceedings — but no AI detector should be the sole basis for an accusation. Best practice is to combine an Imagera scan with at least one independent verification method and human review, and to treat the report as documented forensic evidence supporting a human decision.