From upload to verdict, in five deliberate stages.
Verafai isn't one model guessing "real or fake." It's a pipeline — purpose-built classifiers per modality, a scoring layer, a plain-language explainer, and a retraining loop that never stops learning from what it gets wrong.
Click a stage to see what actually happens inside it.
Every file — video, audio, or image — passes through the same five stages. What changes underneath is which model does the heavy lifting.
Ingest
File or stream enters the system.
Analyze
Modality-specific model scans for artifacts.
Score
Confidence likelihood is computed.
Explain
Flag is translated into plain language.
Retrain
Result feeds the continuous training loop.
The same five stages, three different specialists.
Video, audio, and images each fail in different, specific ways when they're synthetic. Verafai runs a dedicated classifier for each — not a single generic model spread thin.
Video: Frame & Motion
Detects blending seams at face-swap boundaries, unnatural blink cadence, and lighting that doesn't match the scene — frame by frame, then aggregated across the clip.
CNN + TransformerAudio: Spectral Signature
Voice-cloning tools leave behind pitch and prosody artifacts invisible to the ear but visible in a spectrogram. Verafai's classifier reads the frequency domain directly.
Spectral CNNImage: Frequency Domain
Diffusion and GAN models leave a statistical fingerprint in the frequency domain that's invisible to the human eye but consistent across a generator family.
GAN FingerprintingDetectors go stale. Verafai's retraining loop is the whole point.
Every confirmed fake and every confirmed real sample gets fed back into training. When a new generator family appears, Verafai's target is to have a retrained, redeployed model within 14 days — not the months a static detector takes to fall behind.
Every analysis — confirmed or disputed — becomes a labeled training sample.
+Models are retrained on a rolling basis as new generator families are identified.
+Every retrained model is benchmarked against the full library of known generator families before it ships.
+The updated model replaces the previous version in production with zero downtime.
+Questions we get from every pilot team.
Ready to see it work on your own media?
Join the pilot waitlist or request a live walkthrough with the founding team.